Podcasts about Ethernet

Kick off Mac Geek Gab 1157 with a rapid-fire round of Quick Tips: find the new Siri option hiding in your Camera app, tame Archive Utility’s settings to streamline your workflow, and turn your scanner and printer into a DIY copier over FTP. You’ll also learn where macOS quietly stashes your FileVault recovery key (hint: check Passwords) and how to wrangle your files so AI can actually help you, including a listener who let AI sort an entire TV library and another who put Claude Cowork to work as his personal Mac updater. Then it’s time for Apple’s big hardware drop: the M6 and M5 Ultra arrive in a new Mac mini and Mac Studio, with 2.5-gig Ethernet as the buried lede. Your two geeks and a pilot break down what’s worth your money, tackle Apple’s storage pricing problem with Thunderbolt and USB alternatives that won’t drain your wallet, and answer whether you can move your home folder to an external drive to dodge Apple’s SSD tax (spoiler: you can, and maybe it’s time we start endorsing this?). Stick around for Cool Stuff Found: a robot mower that behaves even without fences, a smarter way to peek inside archives, and a slick DMG utility. Press play, get your answers before your gear gets you… and whatever you do, Don’t Get Caught! 00:00:00 Mac Geek Gab 1157 for Monday, August 31st, 2026 August 31st: World Distance Learning Day MGG Monthly Giveaway – Win a license to Yoink, Screenfloat, Transloader, or DeskMat from Eternal Storms The MGG Merch Store is Live! Quick Tips 00:00:01 Pilot Pete-QT-New “Siri” option in Camera 00:02:43 Allan-QT-Use Archive Utility’s Settings to streamline your workflow – From Stephen Robles 00:05:21 Jim-QT-Connect your scanner and printer via FTP to make a copier Also scan to USB for speed 00:08:26 KiwiGraham-QT-FileVault Recovery Key is in Passwords 00:13:10 Beware the stump grinders Call Before You Dig 00:14:05 Eric-CarPlay brings on an ##infinite Siri Loop 00:19:02 Tux-QT-Managing Files for use with Siri AI 00:22:01 Michael-AI Helped me sort my TV Episodes Build Me a Handoff Prompt 00:26:21 Ottograph is coming! 00:32:00 Steve-QT-1156-Using Claude CoWork to Play Mac Updater Sponsors 00:35:34 SPONSOR: Helix Sleep makes premium mattresses and bedding that are customized to fit your personal needs, and conveniently shipped to your door. Go to https://helixsleep.com/MGG for 27% Off Sitewide. 00:37:14 SPONSOR: Proton VPN. Unlike most VPNs, Proton is backed by strong European privacy laws, and, right now, Proton VPN is offering our listeners 70% off a two year plan when you go to https://ProtonVPN.com/MGG 00:38:38 SPONSOR: BBEdit, the power tool for text from Bare Bones Software; now with integrated Notebooks and extended language support. Your Questions Answered – feedback@macgeekgab.com 00:39:46 New M6, M5Ultra New Mac mini M6 and M5 Pro New Mac Studio with M5 Max and M5 Ultra Ian-All with 2.5GB Ethernet as standard! 00:48:12 Pensacola Craig-Apple has a pricing problem Thunderbolt Drives SABRENT Rocket XTRM 5 2GB (TB5) – $400 SABRENT Rocket Nano XTRAM (TB3) – $340 USB 10G drives SANDISK 2TB – $290 Lexar 2TB – $270 00:56:51 Josh-Can I move my home folder to an external drive to save on storage costs with a new computer? Cool Stuff Found 01:05:38 PIlot Pete-CSF-Worx Landroid WR320.1 Robot Mower Works with no fences between you and neighbor 01:13:35 ATCCSF-BetterZip Quick Look Generator and Extension 01:16:18 rndoug-csf-EasyDMGapp (also courtesy of Stephen Robles) 01:17:48 MGG 1157 Outtro MGG Monthly Giveaway Bandwidth Provided by CacheFly Pilot Pete's Aviation Podcast: So There I Was (for Aviation Enthusiasts) The Debut Film Podcast – Adam's new podcast! Dave's Business Brain (for Entrepreneurs) and Gig Gab (for Working Musicians) Podcasts MGG Merch is Available! Mac Geek Gab iOS app Mac Geek Gab YouTube Page Mac Geek Gab Live Calendar This Week's MGG Premium Contributors MGG Apple Podcasts Reviews feedback@macgeekgab.com 224-888-GEEK Active MGG Sponsors and Coupon Codes List BackBeat Media Podcast Network

On today's show we discuss Ara's new family room TV and Audio setup and what his plans are for the future. We also read your emails and look at the week's news. News: Marantz Introduces The CINEMA Series 2 Receivers Dolby Vision 2 HDR Is Officially Arriving On Certain Hisense TVs Nearly A Third Of U.S. Homes Get TV Only Via Streaming Peacock hikes prices for fourth straight year, this time by $3 Other: Theatrium - A home Assistant or HomeKit Dashboard for your AppleTV Parts List for Ara's New Home Family Room: Mantel Mount MM815 Motorized Drop Down & Swivel TV Mount - $700 The MM815 is a remote-controlled drop-and-swivel TV mount. Press one button to lower the TV to eye level and rotate it as needed. It comes with an RF remote, is silent, and has two memory presets. Sony BRAVIA 7 II PRO True RGB 4K HDR Google TV with Gemini (Model Number K75XR70PRO)-  $2300 The Sony BRAVIA 7 II PRO is Sony's 2026 mid-range True RGB Mini-LED TV (65 inches and up) with independently driven red, green, and blue backlight LEDs, XR processing, and about 2,000 nits of peak brightness for stronger color volume and contrast than a typical white Mini-LED. It is 4K at 120 Hz, supports Dolby Vision and Dolby Atmos, and has HDMI 2.1 gaming features (VRR, ALLM, 4K/120). Google TV includes Gemini for voice search and recommendations, plus AirPlay 2. The Pro bundle adds a three-year replacement warranty, a backlit rechargeable remote, extra Sony Pictures Core credits, and an enhanced Voice Zoom 3 dialogue mode; picture hardware is the same as the standard 7 II. From RTINGS: Pros: Great black levels with minimal haloing. Bright details in HDR stand out incredibly well. Fantastic SDR brightness helps it overcome glare during the day. Fantastic upscaling and great low-quality content smoothing. Colors are incredibly bright and vivid. Cons: Dark scenes are hard to see during the day. Sonos Arc Ultra - $1099 The Sonos Arc Ultra supports Dolby Atmos that has a 9.1.4 driver layout and Sound Motion woofer to produce wide, height-enhanced surround and stronger bass from a single bar. It connects to a TV with one HDMI eARC cable, supports Wi-Fi, Bluetooth, AirPlay 2, and the rest of the Sonos ecosystem (including optional Sub and rear speakers), and can be controlled with the TV remote, Sonos app, touch controls, Sonos Voice Control, or Alexa. Trueplay tunes the sound to the room, and AI speech enhancement is meant to keep dialogue clear. It is a long, low-profile bar (about 46 inches wide) sold in black or white. Sonos Sub 4 - $899 The Sonos Sub 4 is a wireless subwoofer for Sonos S2 systems, pairing with current soundbars and speakers to add bass down to about 25 Hz. Two inward-facing 5-by-8-inch woofers and dual Class-D amps use a force-canceling layout so the cabinet stays relatively still and the bass stays clean. It has the same slot-style shape as earlier Subs, in black or white matte, and can stand upright or lie on its side. Updates over the Sub 3 include Wi-Fi 6, a faster processor, more memory, lower idle power, and a slightly lighter cabinet; setup is through the Sonos app, with optional Ethernet. Two Sub 4s can be used with the Arc Ultra. Master Bedroom: Sony BRAVIA 3 II 4K HDR LED Google TV (Model Number K55XR30M2) - $798 - The Sony BRAVIA 3 II is Sony's 2026 entry-level 4K LED set, sold from 43 to 100 inches. It uses a direct-lit LED panel, a 120 Hz refresh rate, and the XR processor for color, motion, and upscaling. HDR covers Dolby Vision, HDR10, and HLG. All four HDMI ports are 2.1, with 4K/120, VRR, and ALLM. It runs Google TV with Gemini, AirPlay 2, and Chromecast, plus built-in speakers that pass Dolby Atmos and DTS:X to a soundbar. The picture is good in bright rooms; but as you would expect, the contrast and HDR are limited compared with higher BRAVIA models. If you are going to use this for anything more than casual viewing or you are really critical of picture quality you may want to look elsewhere. But for those on a budget or used in a secondary room, it's a solid TV.

In this classic episode from the Namaste Archive, Cally talks to comedian Mark Simmons about Mums, sons, Ethernet cables, therapy, break-ups, crying, dreams, Zoom comedy, tampons, TikTok, Taekwondo, tour support, bad gigs, good gigs and jokes. Instagram: ⁠@jokeswithmark Get tickets for Cally's Tour Order Cally's Book More about Cally Produced by Mike Hanson for Pod People Productions Music by Jake Yapp Cover design by Jaijo Part of the Auddy Network Learn more about your ad choices. Visit megaphone.fm/adchoices

Brad and Will reconvene this week to talk about their recent adventures in the real world. Brad returns from the Vintage Computer Festival West once again with a recap of this year's highlights, including projects to resurrect everything from classic America Online to obscure software distribution kiosks from 1980s Japan, panels with some fun details about things like the creation of Ethernet and programming heroics on the Atari 2600, plus some Quake deathmatch against today's youth. And Will returns from Los Angeles to talk about object show meetups and the latest from Disneyland, such as high-tech face projection mapping, why the Lightning Lane isn't so Lightning anymore, the joy of trackless dark rides, and more. Notes and links for this episode: https://tinyurl.com/techpod-351-vcfw-disney Support the Pod! Contribute to the Tech Pod Patreon and get access to our booming Discord, a monthly bonus episode, your name in the credits, and other great benefits! You can support the show at: https://patreon.com/techpod

This week, we're looking at Apple's legal challenge against the UK Government over demands to access user data, new EU rules on AI transparency, and the latest leaks surrounding Googlebooks from Dell and Lenovo. We also take a nostalgic trip back to the 1980s with Timex's futuristic Retroware digital watch, plus some very welcome news involving Amiga and Commodore. With Gareth Myles and Ted Salmon Join us on Mewe RSS Link: https://techaddicts.libsyn.com/rss iTunes | YouTube Music | Stitcher | Tunein | Spotify  Amazon | Pocket Casts | Castbox | PodHubUK Feedback Listening to your thoughts about a streamer with Ethernet. Have you thought about the Manhattan Aero? The Aero is the only streaming option that Manhattan do. The others are all freeview/freesat based. An alternative is the Netgem Pleio... Have no direct experience of the Pleio but will say that the Aero runs on the Tivo platform and won't support Now TV. Got 4 of them when I got hacked off with Sky... Phil I have a couple of compact cameras; a samsung and canon, which are being used regularly since 2010.  I rarely use my phone's cameras even for videos. Almost every event and holiday, those two cameras are present. Oh and I still use my moto mod hasselblad camera with my moto z3 play. JTManuel News Apple launches legal challenge against UK government demand to access data EU announces new rules on AI transparency Dell's first Googlebook could borrow one of its most iconic laptop brands Massive leak reveals Lenovo Googlebook 15 ahead of launch Timex's New Retroware Digital Looks Like a Futuristic Watch from the '80s Amiga And Commodore, Back Together (Sort Of) Banters: Knocking out a Quick Bant Planify FreeFileSync - A Review Alternative - Syncthing Bought Report Yeesabella Grip Stickers Bargain Basement: Best UK deals and tech on sale we have spotted NOCO GENIUS2: 2A 6V/12V Smart Battery Charger - £45.34 Was £54.95 XTEINK X3 3.7" Pocket E-Ink eBook Reader £65 from £80 (Qi2 connectivity and magnetic pogo-pin charging - X4 has USB-C but is bigger) Miady 2-Pack 5000mAh Handheld Portable Fan (7-20 Hours Runtime) - £11.99 Was  £23.99 Samsung Galaxy S25 FE 256GB/8GB £499.00 from £699 (sometimes £599) - £99.80/month (5 months) - the 128GB one is £449 MAGICNUC AG2 Mini PC — Ryzen 7 7840HS, 32GB DDR5, 1TB NVMe - £599.00 UGreen 13-in-1 Docking Station £69.99 from £119.99 Main Show URL: http://www.techaddicts.uk | PodHubUK Contact:: gareth@techaddicts.uk | @techaddictsuk Gareth - @garethmyles | Mastodon | Blusky | garethmyles.com | Gareth's Ko-Fi Ted - tedsalmon.com | Ted's PayPal | Mastodon | Ted's Amazon YouTube: Tech Addicts  

RAD's Chief Technology Officer, Ron Insler, explains why Carrier Ethernet is evolving to meet the demands of AI. As AI workloads become more distributed and performance sensitive, deterministic connectivity is becoming essential for AI-ready infrastructure.  How will networking evolve to support the next generation of AI? In this Executives at the Edge episode, host Pascal Menezes... Read More The post Carrier Ethernet: Built for the AI Era appeared first on Mplify Alliance.

Derrière cet accessoire posé à côté de notre clavier se cache une invention qui a révolutionné notre manière d'utiliser les ordinateurs. Cet épisode de la série « Tout comprendre » raconte l'histoire de la souris informatique, explique son fonctionnement et montre pourquoi, malgré les écrans tactiles et l'intelligence artificielle, elle reste un outil irremplaçable.* Épisode soutenu par Frogans : https://www.f2r2.frUne invention qui a changé l'informatiqueAvant la souris, utiliser un ordinateur signifiait taper des lignes de commande au clavier. Dans les années 1960, l'ingénieur Douglas Engelbart imagine une nouvelle façon d'interagir avec les machines grâce à un dispositif permettant de déplacer un curseur à l'écran.Présentée lors de la célèbre « Mother of All Demos » en 1968, cette invention ouvre la voie à l'informatique moderne, bien avant l'arrivée des ordinateurs personnels.De Xerox à AppleLa souris reste longtemps cantonnée aux laboratoires de recherche avant d'être perfectionnée au Xerox PARC, où naissent également l'interface graphique, le réseau Ethernet ou encore l'impression laser.C'est finalement Apple qui démocratise cette innovation avec le Lisa, puis surtout le Macintosh en 1984, en faisant de la souris le symbole d'une informatique plus intuitive et accessible.Comment fonctionne une souris ?Le principe est simple : transformer un mouvement physique en déplacement du pointeur à l'écran.Après les premières souris à roues mécaniques, les modèles à boule dominent pendant plusieurs décennies avant d'être remplacés par les souris optiques puis laser. Aujourd'hui, un capteur analyse en permanence la surface sous la souris afin de calculer ses déplacements avec une grande précision.La souris a-t-elle encore un avenir ?Malgré la généralisation des écrans tactiles, des commandes vocales et des interfaces en réalité virtuelle, la souris conserve un avantage décisif : sa précision.Pour retoucher une image, monter une vidéo, concevoir un document ou utiliser un logiciel professionnel, elle demeure souvent l'outil le plus efficace. Comme le clavier avant elle, la souris ne disparaît pas : elle continue simplement d'évoluer aux côtés des nouvelles interfaces.Hébergé par Audiomeans. Visitez audiomeans.fr/politique-de-confidentialite pour plus d'informations.

Watch the full episode on YouTube:We first covered Baseten last year when DeepSeek mania was at peak hype. Now they have raised a monster $13B round and become one of the new cohort of AI Infra decacorns that are (with Nvidia, Intel, and the semis complex) chief beneficiaries of the Inference Inflection. We return to Baseten at the peak of the 2026 edition of Open Weights debate. Ali has published a viral breakdown of Kimi K3:And since you last saw him, Philip has spoken at AI Engineer and written the definitive book on Inference Engineering spotted all over SF:Three years ago, inference engineering barely existed as a category.Today, it is one of the most critical disciplines in AI. Inference engineering inherently tackles a different question than standard model training: “How do you turn those weights from training into a product that is fast, reliable, and affordable at scale?” Focusing on these creates an entirely new optimization problem.In one recent GLM-5.2 experiment, quantizing more of the model actually preserved its benchmark quality while increasing throughput by 20%, because the errors introduced in different layers could cancel each other out.Inference is no longer just the final step after training. It is becoming its own engineering discipline, with its own research problems, infrastructure, and increasingly specialized roles.In this episode, Baseten's Philip Kiely and Ali Taha join swyx and Vibhu to explain what actually happens after a new open model is released and what it takes to turn “we generated a token” into a fast, reliable, production-ready API.We go deep on cache-aware routing, disaggregated prefill and decode, quantization, speculative decoding, KV-cache movement, model parallelism, GPU kernels, and the race to make frontier models up to 10× faster. Philip and Ali explain why inference optimizations can still produce gains of 20%, 100%, or even 200%; how quantization errors can cancel one another out; why identical weights can behave differently across clusters; and how Baseten grafted a Kimi vision encoder onto GLM-5.2 without changing the underlying language model.The conversation then expands beyond LLMs into NVIDIA Dynamo, mega kernels, Rubin, AI-specific chips, local inference, video generation, diffusion versus autoregressive models, and the enormous compute barrier to generating coherent long-form video. Finally, we explore the convergence of training and inference, continual learning through persistent KV cache, and the emerging loop where models help optimize the infrastructure that runs them.We discuss:* What happens when a 200,000-token request enters an inference system* Cache-aware routing and reusing previously computed KV cache* Why prefill and decode are increasingly handled by different GPUs* When dedicated deployments become cheaper and more reliable than shared APIs* How speculative decoding uses a smaller model to accelerate a larger one* Tool calling, structured outputs, and what LLMs actually do* What it takes to support a new open model on day zero* Grafting Kimi's vision encoder onto GLM-5.2* Retrofitting inefficient model layers with components from other architectures* Why models sometimes collapse into repeating the same token* How hardware, kernels, and race conditions create nondeterministic failures* Preserving model fidelity while making inference faster* How quantization errors can cancel each other out* Why inference optimizations still deliver gains of 20%, 100%, and 200%* How optimized serving can make a model up to 10× faster* NVIDIA Dynamo, KV-aware routing, and distributed model serving* Speculative decoding the speculative decoder* Why local AI is about making models less dumb while data-center AI is about making them less slow* Tensor, expert, and pipeline parallelism across GPUs* Hardware-aware model design, auto-tuning, and the case against mega kernels* Rubin and why inference is becoming a systems problem* Whether modern GPUs are evolving into programmable AI ASICs* Why enormous models like Kimi K3 require GB300-class hardware* Why open-source video generation still trails Veo, Kling, and other closed models* The quadratic attention bottleneck behind long-form AI video* Autoregressive video, real-time generation, and compounding quality drift* Why future video systems may combine autoregressive and diffusion architectures* Training for inference and inference for training* Continuous post-training, deployment, evaluation, and improvement loops* How GLM-5.2 helped optimize the kernels serving GLM-5.2 itself* Why faster networking could unlock dramatically faster decoding* Continual learning, KV-cache compaction, and persistent model memoryShow Notes* How to build a day-0 API for Kimi K3* 22580: From GPT2 to Kimi3, ExplainedPhilip Kiely* LinkedIn: https://www.linkedin.com/in/philipkiely* X: https://x.com/philipkiely* Inference Engineering: https://www.baseten.co/inference-engineering/Ali Taha* LinkedIn: https://www.linkedin.com/in/aliestaha/* X: https://x.com/waterloointernTimestamps00:00:00 Introduction and the 200K-Token Prompt00:03:18 Dedicated Deployments, Speculative Decoding, and Tool Calling00:11:26 Launching Production-Ready Open Models00:19:06 Model Retrofits, Failure Modes, and Nondeterminism00:28:22 Quantization and Canceling Errors00:32:15 The Race to 10× Faster Inference00:40:48 Dynamo, Speculation, and Local vs. Data-Center AI00:50:18 Model Parallelism, Auto-Tuning, and Mega Kernels01:00:55 Rubin, GPUs vs. ASICs, and Custom AI Chips01:10:03 Giant Models and the Limits of GPU Memory01:12:42 AI Video, Quadratic Attention, and Autoregressive Generation01:21:47 Audio, Images, and Diffusion Models01:27:32 Training, Self-Optimizing Models, and Continual Learning01:40:06 Closing ThoughtsTranscriptIntroduction: Baseten, Waterloo Intern, and Inference EngineeringSwyx [00:00:00]: Okay, we're here in the studio with Philip, old friend from Inference Engineering, the book, as well as Baseten and everything that you've done, you and I have done before, as well as Ali. Welcome.Ali [00:00:15]: Pleasure to meet you.Swyx [00:00:15]: Waterloo intern.Ali [00:00:16]: Waterloo intern, always.Swyx [00:00:17]: When did you get “Waterloo intern” as a handle?Ali [00:00:19]: As a handle? Oh.Ali [00:00:20]: I think the rebranding happened mid-March. When I saw it was open, I was like, “I have to take it. Up for grabs.”Philip [00:00:26]: The problem is that Ali is really good at his job and is not gonna be an intern much longer.Philip [00:00:30]: So we have to figure out who's gonna get the handle.Ali [00:00:33]: Well, I'll pass the torch over to the next intern.Swyx [00:00:34]: Oh, okay. It can be, like, you just pass it to another Waterloo grad.Ali [00:00:37]: To another Waterloo intern. No, bruh.Philip [00:00:39]: Yeah.Ali [00:00:39]: Intern.Swyx [00:00:40]: Intern, yeah.Ali [00:00:40]: And no.Philip [00:00:41]: You gotta get an intern from Waterloo.Ali [00:00:42]: Yeah, I've gotta get an intern from Waterloo.Swyx [00:00:44]: Right.Ali [00:00:44]: But they have to follow the path.Swyx [00:00:45]: Oh, it could, but it could come from Baseten, so it's like whoever Baseten gets from Waterloo.Ali [00:00:48]: Right.Swyx [00:00:49]: Has the title of Waterloo.Ali [00:00:50]: It stays in the ecosystem.Philip [00:00:51]: Exactly.Ali [00:00:52]: Halfway through the internship, you either get it or you're out.Philip [00:00:55]: You should also do, like, a big graduation ceremony where you change the handle.Ali [00:00:59]: Just say it.Philip [00:00:59]: For everybody.Swyx [00:01:00]: You guys are good at ceremonies, clearly. We had a nice launch of the book, very successful. But before we get into all that, I wanna start off with a fun question for you. Okay, you're an expert inference engineer. What happens when I send a long query, say two hundred thousand tokens into Baseten's inference? What's the process of query through GPU model routing, balancing, all that? What is all the stuff that we don't think about?Long Context Requests, KV Cache, and Cache-Aware RoutingPhilip [00:01:26]: With a long query specifically, the first thing that I'm gonna ask is, “Have you sent me this query before, or at least part of it?” and I really hope you have, because it's gonna be a lot easier for me and a lot cheaper for you. So the first thing that we're gonna look at is some cache-aware routing, where we're going to see, we probably have a number of instances, a number of replicas up serving whatever model you're hitting. We want to send this one to something with, number one, available prefill workers, and number two, ideally some cached input already there so that we can skip prefill on at least part of these two hundred thousand tokens. If you're doing two hundred thousand tokens, it's probably coding or a multi-turn agent or something where you would expect to have that cached. If you don't, we're gonna have to send it to a prefill worker. We've at least on certain models disaggregated prefill and decode, so you're going to have one set of GPUs that's solely going to process the input, create the KV cache, and get you your first token, and then that's going to be passed over to a separate set of GPUs, which is going to run decode. We're going to iteratively make those tokens. We're probably going to have some speculator model in front of that. I'm going to assume that you're doing coding, and because of that, our speculator model, which assumes you're doing coding, is gonna have a high draft token acceptance rate. If I'm wrong and you're asking me to summarize every Harry Potter book, it's gonna be slower. And then we stream that output to you and account for it, charge you, a couple of pennies and say, “Hey, would you like to send another one?”Swyx [00:03:04]: Except Baseten doesn't charge by pennies.Philip [00:03:07]: Well, yeah, we charge. I'm assuming that we're talking about the public model APIs. If you are setting up a dedicated deployment, then yeah, it's not pennies.Public APIs vs. Dedicated DeploymentsSwyx [00:03:18]: Yeah, one of the key differentiators when I was talking with Baseten initially was that people who want very high volume just need to rent by the box, ‘cause then it's up to you to figure out how to saturate the box.Ali [00:03:31]: And more often than not, it's, like, way cheaper if you're pushing, like, millions of tokens per hour, if you just pay per hour instead of pay per token.Philip [00:03:37]: Yeah, they do. I think that we've increasingly seen a lot of demand for the pay per token APIs, just because everyone wants to try open models, and then once they find a use case that's really sticky, then they move over to dedicated.Swyx [00:03:51]: Is there a best practice on when it's time to swap over?Philip [00:03:54]: Couple reasons. Yeah, reliability, that's a big one, right?Ali [00:03:57]: Like, if they have a very specific use case, they want you to train something specifically for them, like they want their own spec dec, for instance, for their own traffic.Swyx [00:04:04]: Spec dec is speculative decoding.Speculative Decoding and Custom SpeculatorsAli [00:04:05]: Speculative decoding, yeah.Swyx [00:04:07]: You have to explain.Ali [00:04:07]: Sorry. Like, speculative decoding is like, if you have a huge model, right? And so the model is going to be generating one token at a time every single turn, every single forward pass. So we attach, like, this little, like, parasite, like this layer that goes on top of the model, and this model just has to predict. It does three very fast autoregressive forward passes, and it will predict, like, three certain tokens, and then you do one forward stage over the entire original model in order to see if those predictions were correct or not, and then you accept them or you reject them. Now, this draft model is traffic specific, so if you, like, Philip said, if you're summarizing Harry Potter books, I can train exclusively that draft model on Harry Potter books, and I can guarantee you that I'm gonna accept the three tokens every single time. And so with that case, I increase your decode speed. I wouldn't be able to provide this to you if you're a shared endpointSwyx [00:04:53]: YeahAli [00:04:53]: ‘cause I have no idea if you're doing Harry Potter, if you're doing coding, if you're doing English. We don't know. Also, there was a thing in the book that mentioned that if they really cared about a specific threshold, chapter four, I think. Do you remember that?Philip [00:05:06]: Yeah. The things that you can do is you can set a specific, like, batch sizing, a specific, like, parallelism strategy if you're trying to optimize for, like, throughput versus latency. You can. Maybe a NVFP4 quant doesn't pass your benchmarks and you wanna run a model at higher precision, you could do that. There's just a bunch of reasons why you might wanna have your own endpoint and the biggest one, of course, just being, like, you don't have to deal with someone else doing a hundred million tokens of benchmarking traffic at the endpoint when you happen to be trying to serve your users.Swyx [00:05:40]: Yeah. I think one thing that is. That is a classic journey. Like, it's people is asking the, what happens when you type Google into the browser. Tool calling, is that just, you're generating JSON or is there more complication beyond that?Tool Calling, JSON, and Structured OutputsAli [00:05:58]: Certain customers that we have, they have their own post-trained models, and so they demand a tool calling that's not just, like parse a file or go find the weather. It's something that's very specific and you have to do post-training on this. And if the post-training on the model is not good or if the quantization after the post-training to get the inference to be fast, the model will struggle reading the JSON file and reading the tool calling. But it doesn't require its own like sandbox. It's not like it's going to use that tool calling to like escape a sandbox or like it doesn't have to be contained. It can just be a normal dedicated deployment. The challenge with tool calling more and more seems to be that the companies want certain tool calling which is a very sensitive thing to train. And because you're dealing with all of the JSON outputs, if it doesn't like close the end of the request in a very certain manner, you end up with a model that did the tool calling and like the thinking and so as a result of that, it didn't see the result and just hallucinated the result as it decoded. That seems to be the most challenging thing with tool calling, not really the sandboxes model.Philip [00:06:56]: Yeah, that's a challenge on the training side and then on the inference side, there's work that you can do to scope the possible output. So we published this at this point close to two years ago, the solution to this problem which is you make a state machine and you use that to constrain the output to a specific format. So this is the structured output problem. If you remember backSwyx [00:07:27]: Yeah, the specific grammar is,Philip [00:07:29]: Yeah, exactlySwyx [00:07:30]: GML had this thing.Philip [00:07:31]: Yeah. So it's like the old-school “make sure this is only JSON”, return only JSON orSwyx [00:07:38]: YeahPhilip [00:07:38]: Grandma's gonna die type of prompts.Swyx [00:07:39]: Is it BNF grammar? At some point OpenAI had released a thing that was like, yeah, if you want to constrain your output, write BNF grammar, back as NOR.Philip [00:07:47]: In our inference system, it's just a specified output format. And you get the guarantee that your output's gonna be structured along that format. And so applying that to tool calls can like help cut down on. You can still call the wrong tool or call no tool. It doesn't solve the certainty problem but it at least solves the output structuring problemSwyx [00:08:10]: YeahPhilip [00:08:10]: Within tool calls.Swyx [00:08:12]: And MCP is just another form of tool, right.Philip [00:08:14]: Yeah, exactly.Swyx [00:08:15]: As far as there's no special thing there.Philip [00:08:16]: The thing I'm always like explaining to people is the LLM is not capable of doing anything. It's only capable of making suggestions of what to do and then if those suggestions are formatted in a certain way and applied to a system that knows what to do with them, then an action occurs.Swyx [00:08:32]: Yeah. Part of the fun stuff is, this is solved outside of tool calling too. Like in an agent loop if the output is not correct or you're right, like reasoning, tool calling was done in the reasoning trace, just be like, “Oh, I don't know what to do. Let me just try again.” And it might get there after a few tries. And on your point of training, sometimes this is harder in smaller models, so you don't have the same exact quality outputAli [00:08:56]: Right.Swyx [00:08:57]: When you just swap from a big model, right?Ali [00:08:59]: Yeah. I will say that, before, I think we need to go back to inference engineering proper.Ali [00:09:04]: But, I had expected that something would replace JSON because it's hard to stream JSON ‘cause JSON must be complete and you must have open and close brackets and everything. So it's hard to parse something or validate something while it's being streamed. So people invented all sorts of things that are like, I forget the name of some of these alternatives, but it's something like TOML, something like YAML. But JSON seems to be dominant still.Philip [00:09:30]: The JSON outputs aren't that long, right? Like you could have a long-- ‘cause tool calls also contain the arguments in them and perhaps for a certain tool you might pass like a very long argument. But my impression of the median tool call is that it's a relatively small number of tokens, right? So I would expect that speculators are generally fairly good at something as formatted as JSON. And so you would have like a pretty fast decode step there and that the streaming wouldn't be as valuable, but maybe I'm wrong about that.Ali [00:10:02]: I think you're also bounded by the software or that the model is gonna integrate with if the software is built with JSON for the tool calls or if the company that you'- if your customer says that this is how our software works and our tools are interfaced with JSON, you can ask them to like, change their software and say like, “Yeah, this is gonna be better for the model.” but like with the right training shouldn't be that much of a difference. Also more profitable if it outputs more tokens probably.Swyx [00:10:25]: Depends on your business model.Swyx [00:10:27]: It really depends. But I will say that, as a writer with like experience a lot with generated output, I do try to move from text to JSON text which is very long JSON, right? Like there's paragraphs in every field because I'm trying to structure it, right?Philip [00:10:44]: Right.Swyx [00:10:44]: I want you to first make factual statements, then make opinions then make bullet point summaries, have dates, have entity references have your sources for references, all these things. Anyway, so these are things that like I think people who really experiment with structural output have to really care about. But, let's, let's recurse up the stack a little bit. Before we started recording, you mentioned something really cool, which is that there's a lot of engineering that-- inference engineering that goes on when a new model provider releases a new model, right? So let's call it GLM-5.2, Kimi K3. I had previously assumed, especially if it's like, well, GLM 5 to 5.1 to GLM-5.2, like that you've supported them before. Is it that much work?What It Takes to Support a New Open ModelAli [00:11:26]: It's a lot of work.Swyx [00:11:28]: Yeah. Okay. So like, a lot of people, all you guys, right whenever a new model launch like, people rush to say like, “Oh, Hugging Face supports this, Fireworks supports this, Spacetime supports this,” and I'm like, “Yeah, of course we support it.” But what goes into that? What goes intoPhilip [00:11:40]: I think it's more than just support it too, right? It benefits the consumer a lot. Like I think it was with Kimi K2.5 or GLM-5.2 the latest, there was an inference war, right? X provider is at 90 tokens a second. The next day we're at 150. The nextSwyx [00:11:55]: I kinda kicked that off with the GLM-5.2.Swyx [00:11:58]: I wrote a Twitter article about. It got like half a million views,Ali [00:12:02]: Based on being numberSwyx [00:12:03]: YeahAli [00:12:04]: Or it's for something else.Swyx [00:12:05]: Yeah. Which,Ali [00:12:06]: Oh my GodSwyx [00:12:07]: Which then got everyone really excited about, hey, how can we, bend tracks a little bit further and,Philip [00:12:14]: There's a difference between support the model, as in I can make a token out of this model, and support a model, as in I have a production-ready API from this model.Philip [00:12:26]: Getting to the point of I can make a token out of this model is not that hard because generally the, open source inference engines, vLLM, SGLang of the world oftentimes even receive weights ahead of time, maintainers do, or the people making the model merge PRs to ensure support. So you generally can, just get it working on the standard open source stack without too much pain in most cases. The challenge is, every inference company is gonna have own proprietary stack. Some open source components, some in-house stuff. And for any arbitrary model, there's going to be some new stuff. Sometimes you get lucky, like K, two five to two six was, like, pretty similar.Quantization, Speculators, and Production ReadinessAli [00:13:16]: Yeah. It was pure continued post-trainingPhilip [00:13:18]: YeahAli [00:13:18]: If I remember correctly.Philip [00:13:19]: Even in those cases, there's still stuff you have to do. You have to redo the quantization work. You're taking the model from. Generally, these models are not released in NVFP4, and we want them to be in NVFP4 for maximum Blackwell compatibility. So we have to perform that quantization, and, calibrate the quantization to make sure that we're not causing any regression in the model's intelligence. And then we also have to train the speculator, as we've talked about. Generally, we have. We have ZDR, zero data retention on our model APIs, so we don't know exactly the traffic that people are sending us, but we know what's popular. We know that coding use cases are popular. We know that agents, agentic use cases are popular. So we can get public data sets that are representative of that traffic and train general speculators. Now, with speculators today, you need to train the speculator using the base model itself because you're getting hidden states out of the model from running inference on these specific prompts, and that is the training data you use to create the speculator. So there's that process which you need the real model weights for. And then there's of course just the process of, standing up all the infrastructure behind it, loading all this stuff, testing it. And then when there's a new model with a newer architecture, I think that, like, the DeepSeek models tend to be the most challenging as they have, like, the most novel architectural stuff going on, model after model. But every new model has something. Kimi K2 had. Oh, sorry, GLM-5.2 hadAli [00:14:53]: Sparse attention.Philip [00:14:54]: Yeah,Ali [00:14:54]: YeahPhilip [00:14:54]: the DSA.Ali [00:14:55]: Right. Which is brought from DeepSeek.Philip [00:14:57]: Yeah. AndAli [00:14:59]: So you can copy-paste then?Philip [00:15:01]: It kindAli [00:15:01]: I don't know how this works.Philip [00:15:02]: So, like we had to, like, build support for that into our runtime. And you're right, like it is really interesting the way that all of these open source labs borrow from each other. For example, like GLM-5.2 doesn't have vision. So something that, Haley, a guy on our team, if we could take a look at this, he, like, grafted the Kimi vision encoder onto GLM-5.2.Retrofitting Vision into GLM-5.2Ali [00:15:27]: We'll be training the projector.Philip [00:15:28]: Exactly. So if you think about, like, the encoder, there's the encoder, which is the part that looks at the image and turns it into latent information, and then there's the projector which likeAli [00:15:38]: You can say latent space. It's okay.Philip [00:15:41]: And then there's the projector that maps it onto, the model itself, and then there's the model weights. You don't wanna mess with the model weights because you run a chance of making the model dumber at something else for the purpose of giving it vision. So instead, Haley started with just a projector, which is only a handful of millions of parameters.Ali [00:16:02]: That would be, yeah.Philip [00:16:02]: Yeah.Ali [00:16:03]: Can you show the training one?Ali [00:16:04]: Like the way it groksPhilip [00:16:05]: YeahAli [00:16:06]: Very interesting.Philip [00:16:06]: And maybeAli [00:16:07]: That right therePhilip [00:16:07]: Maybe Ali, you should take it from here. You've got a betterAli [00:16:10]: Ooh, double the sandPhilip [00:16:11]: Understanding of this than I do.Ali [00:16:11]: Yeah. You can see, like, he. The way he trained this is really cool. At the beginning, he was training it using just like, “Here's a picture of a mountain. Can you describe what's in this mountain?” And that caused it just like the first, learning walls. Like here you can see this all we're trying to teach it is to translate the encoded. Like it's already taken the encoder from Kimi K. It's taken the image. It'Philip [00:16:31]: Yeah. FrozenAli [00:16:31]: FrozenPhilip [00:16:32]: With adapter.Ali [00:16:32]: Exactly.Philip [00:16:33]: Yeah.Ali [00:16:33]: So the brain is frozen and the eyes are frozen. It's just we're tryingPhilip [00:16:37]: AlignAli [00:16:38]: Interconnect between the eye and the brain, right? So the projector. And so you take the tokens and then he's like, “Oh, can you describe what's in this image?” And he's like, “Oh, it's a mountain,” or it's a person or it's a human, whatever the case is. But that didn't cause complete understanding. So he changed it such that every image was associated with a data set of questions. Like, does this image have a white male? Does this image have birds in the top corner? Does this image have a scientist in it? All of that stuff. And it would have to answer questions correctly. And using not just training on describing an image, but being able to answer question, another question, answer over time. Like you can see the grokking, which is like genuinely insane, that retrofitting vision into a large LLM can learn to that extent. And even for images that it doesn't perform well on, for instance, if you ask it a picture of like Stephen Hawking, “Who is this?” Maybe it doesn't get it, but it will say something like, “This is Albert Einstein.” Like it still understandsPhilip [00:17:25]: Close enoughAli [00:17:26]: That this is a scientist who is a man who has, some significant achievements, all that stuff. So that's like really cool.Philip [00:17:32]: Yeah. So, we've covered Hao Tian before, who the author of the LLaVA paper that did this, a while ago. And I think that's very foundational work for anyone who hasn't done vision work before.Ali [00:17:41]: Same with the CLIP and MetaCLIP, where you go from just captioning to building out questionsPhilip [00:17:47]: RightAli [00:17:47]: Off the image and how much better you can get performance.Philip [00:17:50]: Right. Right. Right. Yeah. But what's, what's so exciting about this is if you look at a model like this. Now, this is a little bit more of a research project. It's not. It got to 56% on MMLU Pro, I think. So not quite frontier. But if you're running this model, you haven't suffered any loss on your GLM-5.2 quality. If you don't have an image, it'll just behave exactly the way it used to. And ultimatelyAli [00:18:14]: Which in the inference code you literally do not include the other part, right?Philip [00:18:18]: Yeah. You would just skip the encoder if you don't have an image input.Ali [00:18:22]: Okay.Philip [00:18:22]: Just confirming.Philip [00:18:23]: YeahAli [00:18:23]: Does it affect a lot on the overall inference side? Like you're not adding much, you're adding a very small vision encoder. These are typically likePhilip [00:18:30]: They're super fineAli [00:18:31]: Less than a billion parameters, right?Philip [00:18:32]: Yeah. It's, - There's a little bit less standardization among vision encodersSwyx [00:18:37]: YeahPhilip [00:18:37]: So the support matrix can be a little bit, sparser. But overall, yeah, it's a pretty, it's a pretty minor component of the overall system. And ultimately what you get out of the system is all of a sudden you have Kimi Vision, GLM weights, and DeepSeek attention all in one model.Open Source Model Grafting and Franken-MergesPhilip [00:18:56]: And that's, I think, a lot of the power and beauty of open source, is that you can take all of these different components and combine them together into a system that's better than anyoneSwyx [00:19:05]: YeahPhilip [00:19:05]: Can be individually.Swyx [00:19:06]: People used to say that you would also do Franken-merges where you would take likePhilip [00:19:10]: YeahSwyx [00:19:10]: Layers from each model.Swyx [00:19:11]: Does anyone do that anymore?Ali [00:19:13]: Well, to your point previously when you were mentioning like, the work that goes into supporting a model when it first comes out, like GLM-5.2 or MiniMax M3 or whatever the case is. Sometimes you do have to like, you do have to switch out some things. Like, for instance, the MiniMax M3 head uses full attention, and with full attention you end up with this like insane bottleneck in spec dec ‘cause you're doing auto-regressive token generation for three tokens, and you're doing this like N squared over all of the tokens that are in your sequence. Your KV cache is like very large because it's not sparse, it's not top K. So we find it better to like, okay, we're gonna replace this, we're gonna replace this layer with a layer from another model that's using like GQA, for instance. And then just with the right training, you can get it to have the same acceptance rate. So it is very possible to retrofit layers from other models and very much needed. If a layer is like inefficient, the training just becomes the challenge, like how do you ensure that you train it properly? Which again to your earlier point is like the mesh between training and inference. As in like you need very good training in order to do fast inference. That's like, I feel like more and more becoming true.Swyx [00:20:21]: Yeah. Anything else on the support side when you say like get it to fully production ready?Loop Detection, Race Conditions, and Non-DeterminismPhilip [00:20:26]: Yeah. I think that there's also a question of just, we can test a model to a pretty extensive degree, but we're trying to get it out quickly and then you see a bunch of other people test it and you get interesting results. There was an issue with, GLM briefly where we had some like mode collapses where it would just output the same token over and over again for certain prompts on certain temperatures. Like once you expose an endpoint to the real world, there's going to be, so many more varieties of things given to it that you're able to, discover and patch things. So it's not just a, day zero process, it's then like for the first week, for the first month, if a model remains popular, like how do you both fix bugs and then continue to push the envelope on performance?Ali [00:21:21]: What do you mean you don't want your model outputting S?Swyx [00:21:24]: Is there loop detection on that stuff, by the way? It still happens like quite a lot, which is surprising.Ali [00:21:30]: We have like we, in our endpoint, like if a model was to output the same token like four plus times, we just cut the generation. We say like, “Oh, sorry, this-- Like try again,” or like we will reprocess the request. ‘Cause we know then, like if it, like if, yeah, it's four times the same token, it's probably collapsed.Swyx [00:21:45]: Yeah. Is there a way to opt out in case I really want that?Ali [00:21:48]: You want that?Ali [00:21:50]: I think there's a way that we have to handle it. I'm not exactly certain, but I feel like in certain models, like when they output something like you can imagine, like a table for instance, and so they want, they wanna draw like 12 dashes and 12 dashes. Yeah, I think there's a way for that to happen. I think we only do it on certain tokens. Like we exclude certain special characters.Swyx [00:22:07]: Yeah.Ali [00:22:07]: So we only do it on like certain like S is the most common almost. GLM-5.2Swyx [00:22:11]: OhAli [00:22:11]: And I think it was DSV 4 as well. Like you'd just have like looping issues where like you literallySwyx [00:22:17]: ItAli [00:22:17]: Just have like S.Swyx [00:22:18]: Yeah. Is there a special, something special about S? No, just randomlyAli [00:22:21]: It just seems to be the one token involved.Swyx [00:22:23]: Yeah. And it'Philip [00:22:24]: Is thereSwyx [00:22:24]: And it's only temperature 0Ali [00:22:27]: NoSwyx [00:22:27]: Even at other temperaturesAli [00:22:27]: Even at like 0.9 or whatever, it will still, it will still collapse.Swyx [00:22:30]: That's weird, right?Ali [00:22:30]: It's, it is an inference problem to be honest, like a software problem. Like oftentimes, the image you run will-- like NVIDIA will release an image for instance, and if we will upstream the changes from their latest TensorRT-LLM image into our stack, we'll find that it fixes it. Or oftentimes this will only happen in an inference engine that you're using like SGLang. But if you were to switch to vLLM, that isn't the case. So it seems to be like an extremely like deterministic software issue and not really a model issue. It's not like a weights problem. Like I'- we'll say like, “Oh, it's a problem with the quant. We did PTQ wrong,” right? But that isn't, that doesn't make sense because the same weights used with a different inference engine does not repeat the problem. And sometimes it's, the kernels that are being used in the backend have like these very subtle sometimes race conditions, where if you were to use this model hosted on one cluster, you will never get this problem.Swyx [00:23:19]: Oh my God.Ali [00:23:19]: But if you host it on a different cluster, you will. And the reason is the KV cache transfer from a node to node in that one cluster is using a slower interconnect than the node to node in another cluster. So that exposes the race, whereas in another cluster it doesn't. So then you end up just like, okay, this model is not gonna be hosted on this cluster. We're gonna host it on, another cluster because that cluster exposed that problem. But then it ends up with like, okay, is it the software? Is it the model weights or is it the hardware?Swyx [00:23:42]: There is a thing about this with temperature 0 still not being deterministic, right?Ali [00:23:46]: Right.Swyx [00:23:46]: Mostly because of hardware. Even at temperature 0 same model, you won't always get the same output.Swyx [00:23:52]: Even-- But I'm surprised by the race condition one because, I thought PyTorch was a graph that like guarantees that you at least, execute things in the right order.Ali [00:24:02]: Well, yeah, true. Like I'm not, I'm not saying that there is. Like well, you have things like PTL optimizations where like you can start a kernel before the end of the previous kernel, and that's like ‘cause you want to do that because there'sSwyx [00:24:12]: It's like pipeliningAli [00:24:12]: Expense. Exactly.Swyx [00:24:13]: Yeah.Ali [00:24:13]: But it'- But you don't do it cleanly. Like you overlap a little bit of the execution. No, it is very possible that the kernel itself, like that one block that is supposed to be running in this instance of time, that kernel itself has a race condition. For instance, like a missing barrier. Like often if you're designing a kernel and you want it to make it to be very fast, if you don't test it extensively, you'll, you'll have certain threads access data points from registers before they've been written to by other threadsSwyx [00:24:36]: YeahAli [00:24:36]: For example, because like your barrier is wrong or your synchronization was wrong. But yeah, like the testing itself is very difficult in those like, andSwyx [00:24:42]: And there's no like borrow checkerAli [00:24:45]: What does that mean?Swyx [00:24:46]: Like Rust. Like the. If you're trying to have like memory safety It sounds like a comparable problem.Ali [00:24:52]: Well, yes, but you're working in CUDA, right, NVIDIA GPUs. Like- You just need a higher level language like modular Maybe that's what modular is supposed to do. I don't know.Quantization Quality and Vendor FidelityVibhu [00:25:00]: How do you see keeping quality of the model? So you talked about all these steps of, okay, you gotta do quantization, train your own speculative decoderAli [00:25:07]: RightVibhu [00:25:07]: Run on different hardware. Looking at other model providers, okay, you kicked off a inference speed race on the consumer end. What goes into keeping quality the same across them, right? Sure, you can run benchmarksAli [00:25:22]: YeahVibhu [00:25:22]: But, like, how do you determine how much quantization are there standards? What goes intoPhilip [00:25:27]: There's a few things on quality. Most inference optimizations are lossless. KV caching, for example. You are just recomputing or preventing recomputing the same values. Speculation, of course, if a draft token is wrong, it gets rejected. The main lossy optimization is quantization. And that really comes down to, number one, data format, number two, which parts of the model you choose to quantize, which layers, and number three, like doing a lot of calibration on the quantized weights, to ensure that you're preserving all the outliers. There's other tricks that you can do, though. A big one is long context, ‘cause one thing you asked at, right at the beginning is, “Oh, what's gonna happen if I send a 200,000 token request in?” So with a long input sequence, you need to, store a lot more information. You need to process a lot more tokens. And so even if a model has a context of a certain length, you might, as an inference provider, choose to build an API with a shorter context length, and of course a full length one as well. Because if someone doesn't need the full million token context, for example, you can get them better performance. I don't know if that's exactly like quality of the model. The way that I think about quality is to what degree are we faithfully serving the original model? If you think of a golden implementation of a model that performs exactly the way the model is designed to perform, I think of quality as how close are we getting to that, 100% fidelity of the model.Philip [00:27:13]: You can also, of course, think about quality from the training side and how do you push yourself past 100%. But when I think about purely inference optimizations, it's getting faster while staying as close to that 100% fidelity mark as possible. And certainly our standard internally is that, like you should not be able to tell the difference between our API and a, official API. I think Kimi in particular does a good job of vendor benchmarking hereAli [00:27:41]: YesPhilip [00:27:41]: Where they haveAli [00:27:42]: They released an actual vendor benchmark.Philip [00:27:43]: Exactly, yeah.Ali [00:27:44]: ‘Cause they accused, some people, Amazon? There was some provider that was not doing very well on Kimi's benchmark.Philip [00:27:50]: Yeah.Philip [00:27:51]: So, with Reflect we probablyVibhu [00:27:52]: This was a long time ago, right?Philip [00:27:54]: No.Ali [00:27:54]: Yeah, like threeVibhu [00:27:55]: They alsoAli [00:27:55]: Four, five months agoVibhu [00:27:57]: This also happened with, I don't remember which model, but they pulled out quite a few, and then they started a whole chart about this. It might have beenPhilip [00:28:03]: Kimi Vendor Verifier.Ali [00:28:04]: Yeah.Philip [00:28:05]: Yeah.Ali [00:28:05]: Yeah, ‘cause you, ‘cause you'd be pissed, right? Like if you'Philip [00:28:07]: Yeah.Ali [00:28:07]: If like if I'm a consumer and I'm using like Amazon's endpoint for instance, and I've used Kimi and I'm like, “Oh my God, like this is bad,” I'm not gonna say, “Oh, Amazon quantized the model in a bad way.” I'm gonna say, “Oh, Kimi sucks.” Right?Philip [00:28:17]: Yeah.Ali [00:28:17]: So it seems like that makes sense.Philip [00:28:19]: Yeah, they care. They care.Vibhu [00:28:21]: Justifiably.Ali [00:28:21]: Yeah, justifiably.Vibhu [00:28:22]: This is probably a stupid question, but just checking, has anything improved from main quantization?Philip [00:28:28]: Yeah.Vibhu [00:28:28]: Like, is quantization always strictly worse?Ali [00:28:30]: Well technicallyVibhu [00:28:32]: NoAli [00:28:32]: It's a lossy. QuantizationPhilip [00:28:33]: YeahAli [00:28:33]: Is a lossy, it's a lossy implementation.Philip [00:28:36]: Speed improvesVibhu [00:28:36]: Speed improves.Ali [00:28:37]: It the number, likeVibhu [00:28:38]: No, I' always look for inverse scaling laws.Philip [00:28:40]: Yeah.Ali [00:28:40]: Yeah.Vibhu [00:28:40]: This is something I learned from Noam Brown, where like things that normally act in one direction sometimes do.Philip [00:28:45]: Well, technically when you run a benchmark, because these models are deterministic, sometimes your,Ali [00:28:52]: YeahPhilip [00:28:52]: NVFP4 quant is like, two basis points higher than yourAli [00:28:56]: No, it's noise. It's noise.Philip [00:28:57]: Yeah, exactly. I'm like, yeah, it's, it's within. That's why I always say within margin of error.Philip [00:29:01]: And I stopped saying that because everyone assumes that what is, well, within some margin of error, we're barely inside of that to the worst, so we're saying. But yeah, sometimes it's just like, gives you a higher output score. But like Ali said, that's noise. To my knowledge, you're not necessarily making the results better. You're just trying to, again, like keep your fidelity as close to 100% to the original model.Layer Selection, KL Divergence, and Better QuantizationAli [00:29:27]: There is, to your point, research that we did on MP. I don't know if you are able to pullPhilip [00:29:31]: YeahAli [00:29:32]: A tweet we did. One of our research interns, Joshua, I think it's a tweet on how we have 20% better quantized GLM-5.2 than NVIDIA. Essentially what we found throughout like this month research is, okay, quantization is a lossy. It's. You're compressing the data from, occupying 16 bits to occupying, four bits, for instance. And so you're losing some information, and you're trying to minimize that. And so when I say that I'm gonna quantize the model, my job becomes how do I find the layers that I can quantize, and how to find the layers to not. For instance, with image models, I don't quantize modulation layers, and I don't quantize out projections because those two are. Like out projection is what you see as the user. Modulation is what the model sees or understands. Right, exactly. And so to his paper, do you have the. It doesn't have the. Yeah. It's a long paper. I don't know if I can findVibhu [00:30:25]: If there's a part to search or it's probably in the thread.Ali [00:30:28]: It's probably in the thread.Vibhu [00:30:29]: Yeah.Ali [00:30:29]: But the long and the short is it is very possible that quantizing more of the model makes the results. Like if I have a model that I quantize layers one, five, and 10, and another model where I only quantize layers one and It is possible that the model in which I quantized more information is going to perform better because the quantization errors have canceled out. And so what Joshua showed in his mathematical proof where he had like a verifier in, is that you can predict which layers are going to have quantization errors that will cancel out with each other, and you choose to quantize those layers. And so the result of doing this mathematical quantization is you end up with a model that's 20% more quantized than another provider, so you get 20% more throughput of it because there's more layers than running an NVFP4, and your quality is better than that other quant because the layers that you chose to quantize have their errors cancel out, like one layer skewed to the right one layer skewed to the left, one layer skewed to the right. Your final logits distribution is more similar to the original distribution of the model, so you have better fidelity. And so the way we proved this was with KL divergence. So instead of just scoring on the benchmarks, we scored the KL divergence between the logit distribution of the quantized model and the logit distribution of the original full precision model, and we showed that with this technique we get. If your probability distribution on the logits which token it wants to select is more of the same as the original model, you're probably gonna end up staying true to the original model. So yeah, so it seems like previously before this, it seemed like the industry was, well, the more you quantize, the worse it's gonna be, ‘cause the more loss you introduce. That's not exactly, not necessarily true. So yeah, doesn't improve it, but can cancel out.Philip [00:31:57]: I think it might be this, but reminds me a good bit about pruning where you can prune off certain layers.Philip [00:32:03]: But very interesting. Didn't know this was a whole paper you guys put out.Ali [00:32:06]: It's. Fun fact, it was originally 72 pages, this paper, and then we decidedPhilip [00:32:11]: WowAli [00:32:11]: We can't tell. We couldn't release it. So it's now 45.Swyx [00:32:15]: Still 39 pages, so very substantive. We talked about evals and all these things and, like what's possible in terms of speedup? Like it's like probably like the numberInference Speedups and BenchmarkingSwyx [00:32:25]: Thing that people do wanna care about, and it's something that you wrote about in your post. Like official API is 70 tokens per second, and you push it up to 90. Is that like a normal thing?Philip [00:32:36]: So what's cool about working in inference, the reason that I think inference is going to be a useful place to do engineering for a long time, is that if you look at highly optimized domains like, say, finance, if you're in finance, you measure how much better you got in basis points. It's like, “Oh, I got five basis points better, like twentieth of 1% better,” that's huge news because everything is so optimized. When we publish optimizations, it's 20%, it's 100% it's 200%. So there's still probably like a lot further to go, honestly. Like you'll, you'll know that inference is pretty much solved when researchers start publishing about how they got 1% faster at something.Swyx [00:33:19]: Which by the way, because I am from the finance background, in the ‘70s, that was the margin at the time. When you did quantitative finance research, you would findAli [00:33:27]: And like 20%, tens of percent.Swyx [00:33:29]: That's. Yes.Philip [00:33:29]: Yeah.Swyx [00:33:30]: And now it'Philip [00:33:31]: Tiny fractionsSwyx [00:33:32]: For those people interested, look up Andrew Lo's paper. He had a really interesting illustration of quant, stat arb, distribution, narrowing down from like those kinds of 20% differences in the ‘70s, down to nothing today, which is very cool.Philip [00:33:48]: Exactly, and we're at the beginning of the same type of thing. Now benchmarking is hard. I think anyone will tell you that, and benchmarking provider speeds is hard because there's so many variables that go into it. What hardware are you using? How much load do you have on the system? What's the exact nature of the prompts and input and output sequence lengths? All that stuff. But overall, when you start stacking these improvements, you're looking at multiples. You can look at it. The most common form, of course, is TPS, tokens per second, which is bad naming by us in the industry, ‘cause there's two tokens per second. There's tokens per second, the throughput number, and the latency number.Ali [00:34:31]: TTMT, yeah.Philip [00:34:32]: Like total tokens per second out of the, out of the GPU as a throughput number. Most people only care about tokens per second as the latency number, which we should call ITL, intertoken latency, but we don't.Philip [00:34:44]: Anyway, so you can imagine a standard API without many optimizations for a 1 trillion parameter model operating somewhere in the 30 to 50 tokens per second range for reasonable traffic profile. And we generally see the goal of, pushing to 10X that. But, not necessarily day zero, but by stacking enough optimizations, if you have, say like four optimizations, each of which doubles performance. Or sorry, three optimizations, each of which doubles performance, then you stack that up, that's an 8X gain. That's the order of magnitude that we're working with in this space. We're trying to make things substantially faster, not just go from like 70 to 90.Swyx [00:35:38]: Are you saying you've. You have done that?Philip [00:35:40]: So let's say you have as a reasonable baseline, 30 or 40 tokens per second. You can achieve 10X that. So like on GLM-5.2, if you run it unquantized, perhaps on H100s even, and you're just using an off-the-shelf inference engine with no particular optimizations, no speculator, nothing extra around like KV routing, no disaggregation, you're, you're probably, yeah, looking at that like 30 to 40. You think that's like a reasonable baseline?Swyx [00:36:12]: Right. Right.Philip [00:36:12]: To get to something like 10X, there's a lot of trade-offs that you're making. If we're running at more like a 300, 400 tokens per second range, you are using the best hardware possible. You have a optimized speculator. You have done all of your quantization work. You are Seeing a pretty high cache hit rate. You are running with a reasonably small batch size and a parallelism configuration that is tuned for latency versus throughput, but it is possible. So the spreads that you see if you, like, go on artificial analysis or you go on OpenRouter and you look at, the worst provider to the best provider, oftentimes can hit that range. 10X is of course very aggressive. It's oftentimes maybe more of a four to six times improvement. But that's the performance that makes us really excited, is when we can get these huge gains, not just go from 70 to 90 tokens.Stacking Optimizations: NVFP4, Speculation, and DisaggregationAli [00:37:19]: It's also, like, hardware dependent. Like, ifPhilip [00:37:20]: YeahAli [00:37:20]: If you have a thing where you're serving it on just, like, a node of H100s and then you throw, like, you shard the model across, like, four nodes of B200s. Like, you can definitely increase the speed with just throwing more hardware at it. Like, normalizing for the same exact hardware and the same number of GPUs.Philip [00:37:35]: Yeah. Then you're looking at, like, a two to 4X improvementAli [00:37:38]: Right. RightPhilip [00:37:38]: Depending on the inference optimizations. So yeah, it's. Some of it's, what's the call, and some of it's who's the driver.Vibhu [00:37:46]: If you break down the two to 4X, say the example is run GLM-5.2Ali [00:37:51]: YeahVibhu [00:37:51]: On B200sAli [00:37:53]: YeahVibhu [00:37:53]: Single node, right? What's, like, the cost trade-off for effort to get, like, the last bit of juice out versus what should people just think of, right?Ali [00:38:01]: Spectre quantization. Yeah.Vibhu [00:38:03]: Spectre quantization.Ali [00:38:04]: That's, that's, that's like 95%. LikeVibhu [00:38:06]: And how far does that get you? And how easy is that for the average person to do? So say right I wanna throw the weights of GLM-5.2 on a node of B200s, how easy is it to find speculative decoder- decoder model or already quantized model? How much work goes into it?Philip [00:38:23]: If you're doing it up front, it's quite a lot of work. If you're doing it today, there's going to be people who have published things that you can just, you can just grab some NVFP4 weights. You can grab a speculator. Yeah, if we're thinking about, like, what are the 2Xs we're stacking, going from, BF16 to NVFP4 is, it's not quite a 2X, right? It's like. I think it's about, like, 30 to 40%, from 16 to 8, and then another 30 to 40% multiplied from, 8 to 4. So that doesn't quite get you a 2X, but, like, roughly a 2X. Speculator, roughly a 2X. Disagg on top of that if you're able to get enough hardware and put enough traffic through it, another roughly a 2X. And then you add in some, double-digit percent increase from having just a better runtime with, the latest kernels and stuff behind it. And that's how it stacks up.Ali [00:39:21]: YeahPhilip [00:39:21]: So building each of those, like, building the, quantized weights is, for someone who really knows what they're doing, hours to days of work. Building the speculator, again, like, hours to days of work. And the, disagg setup, hours to days. Well okay, but like once you haveAli [00:39:39]: Once set up. Once set up. YeahPhilip [00:39:40]: Yeah, getting disagg working for the first time, I'm saying, of course, is very difficult.Philip [00:39:44]: The marginal implementationAli [00:39:48]: Like, if you're just grabbing, like if you are a person, like just a normal consumer who has access to, like, a node of B200s and you're wondering, “How can I just host it myself?” You don't need to quantize the model yourself. There's always gonna be, like, an open source quantized checkpoint. NVIDIA's gonna push one out if no one else does. You. Usually, the providers will have their own spec dec that they've trained as well. You don't need to train your own spec dec. You can just use that as well.Philip [00:40:09]: Yeah. Like, GLM-5.2 has its own MTP.Ali [00:40:13]: Right. Right.Vibhu [00:40:14]: What's multi token prediction?Philip [00:40:15]: Yes.Ali [00:40:16]: I'm justVibhu [00:40:16]: Can you explain that?Ali [00:40:16]: I'm just an expert.Ali [00:40:18]: I can do it for you in case I get it wrong?Vibhu [00:40:20]: No.Vibhu [00:40:21]: Yeah, you should correct if we're wrong, but their multi-token prediction can be used for self-speculative decoding.Ali [00:40:27]: I'm not sure. I'm not gonna correct that.Vibhu [00:40:28]: Okay. I'm semi-confident in thatAli [00:40:30]: Okay. YeahVibhu [00:40:30]: But someone can check. But it's useful to paint the story of, okay, not just the average person, but say a company wants to switch from serverless inference I wanna throw this up on. I wanna rent some GPUs, throw it up. These are the steps you take to do significantly faster than just put it behind vLLM.Ali [00:40:48]: Right.Vibhu [00:40:49]: I was waiting for a mention of Dynamo.Vibhu [00:40:51]: I feel like, that's supposed to be the baseline that you measure against.Dynamo, KV Routing, and Disaggregation ToolkitsPhilip [00:40:55]: I would think of Dynamo as less of a box system and more of a toolkit for building with. So when we talk about doing aware routing, when we talk about doing KV offloading, when we talk about doing, PD disaggregation, Dynamo fundamentally is. By the way, Dynamo is an open source library from NVIDIA.Ali [00:41:17]: We've done a pod with KylePhilip [00:41:18]: OkayAli [00:41:19]: Kyle Cranin.Philip [00:41:19]: Cool. So then your listeners know then that it supports all the different inference frameworks. And it is multi hardware, which is interesting.Ali [00:41:28]: But it's just a router, it's not like an optimizer layer.Philip [00:41:30]: Yeah. All it does, like, what Dynamo is good at, it is a library for moving information around your cluster, around your hardware. So if you have, KV cache on one place and you need it to be somewhere else, Dynamo coordinates NIXL for you to move that around.Philip [00:41:49]: That doesn't mean that, like, out of the box, you just say, “Pip install Dynamo,” and then you get, like, a massive performance speed up. It's more of a developer toolkit.Ali [00:42:01]: Yeah. I would have said it would. It comes with a set of defaults that you can then swap out.Philip [00:42:06]: It does. If the industry at large, I think, was, like, rolling out all of these deployments, standard, then I think it would be, like, a credible baseline. But, we've got to, we've got to benchmark against, like, what we're seeing in the wild.Speculative Decoding Methods: Medusa, EAGLE, n-Gram, and Spec-SpecVibhu [00:42:23]: I did wanna talk a little bit more about PD disagg, because that is probably, like, number three after quantized and speculative decoding. In your book though, I was just gonna pull out the book.Philip [00:42:31]: Yeah.Vibhu [00:42:32]: Like section 522 on Medusa, 523 on EAGLEPhilip [00:42:35]: YeahVibhu [00:42:36]: 524 on gram.Philip [00:42:37]: It's 55, would be disaggregationAli [00:42:42]: Yeah. Well, no, I just wanted to dwell a little bitPhilip [00:42:44]: YeahAli [00:42:44]: The other. Like, so what do you choose to include? What do you choose to not to include? Because there was all these other techniques.Philip [00:42:51]: Yeah.Ali [00:42:51]: Are these still relevant? Because I think they came out, like, a year and a half ago maybe.Vibhu [00:42:55]: Medusa is quite old.Philip [00:42:56]: Yeah, Medusa's old.Ali [00:42:58]: It was old.Vibhu [00:42:58]: But is it in the book as a good, here'sPhilip [00:43:01]: BaselineVibhu [00:43:01]: Baseline vanilla understand it?Philip [00:43:02]: Like you should know this.Vibhu [00:43:03]: Like I read the paper, I'm like, “ it makes so much sense.”Philip [00:43:05]: Yeah.Philip [00:43:05]: So with the book, I had a couple goals. One was to give people just a working vocabulary for the space as a whole, and the other was to give them some intuition about how each of these techniques works. As I mentioned in my AI Engineer talk, which is the first public addendum to this, the speculation space has moved much faster than everything else. So yeah, even at the time that I wrote the book Medusa, I very much included as a way for people to understand how the space evolved rather than what the most modern technique is. And now of course, there's DFlash, dSpark. There's, there's newer techniques even than EAGLE, although EAGLE is still very commonly used.Ali [00:43:51]: SpecSpecta.Philip [00:43:52]: Yes. Speculative decoding.Vibhu [00:43:54]: What canAli [00:43:56]: Oh, it's a paper by Tri Dao and it's like, it's doing speculative decodingVibhu [00:44:00]: HuhAli [00:44:01]: For the speculative decoder.Philip [00:44:02]: Oh, in spec- oh my God.Ali [00:44:02]: It's literally just an another. It's like, yeah, that's the most simple way to explain it, and it seems like he got trivial speed ups there. But it seems that the complexity with training, it's almost like in our mind at least, it's almost as complex as training GANs. Like it's like a very delicate balance and oftentimes you, it's just but yeah, it's literally speculative decoding on speculative decoding.Vibhu [00:44:21]: Speculative.Ali [00:44:22]: Yeah. We saw this paper.Vibhu [00:44:24]: It's interesting, right?Ali [00:44:24]: Yeah.Vibhu [00:44:24]: I wouldn't even expect it to be very particular to train, I wouldAli [00:44:29]: Right.Vibhu [00:44:29]: The naive part of me is like, okay, train speculative decoder.Ali [00:44:32]: But like, and it makes sense, like the whole idea of speculative decoding is you. It's like, it's like almost like the iPhone auto predict version but for a normal model, right? Like you're just, you're just, generating three tokens and you're like, okay, I'll do prefill on them. And so you save those three turns for your original model. Now your speculative decoder is doing three turns of auto regression, so why not just have an even smaller model?Ali [00:44:53]: The other question there is what are the size of speculators? So say forPhilip [00:44:58]: Right. It's like a billion parameters.Ali [00:45:01]: Like for MiniMax, it's. Yeah. It's like one layer. It's like one 60th of the original model usually.Philip [00:45:06]: Yeah. I think we should do a paper when we get back to the office.Philip [00:45:10]: SpeculativeAli [00:45:11]: SpeculativePhilip [00:45:11]: Decoding.Ali [00:45:13]: No, it's, it does seem like how, when do you stop? But then it also seems like if you're able to train spec-spec decode for instance, right? Like if you're able to have a small model that is accurately predicts what the intermediate speculator is gonna predict, that is able to predict what the original target model's gonna predict, then why not just use that smallest model directly, right?Vibhu [00:45:34]: Yeah. This isAli [00:45:35]: Like it seems likeVibhu [00:45:35]: Adjacent to the routing problem.Ali [00:45:36]: Right.Vibhu [00:45:36]: Yeah.Ali [00:45:36]: Right.Philip [00:45:37]: The thing with speculators is one of the practical constraints on using them is that you do have to run a small model on the same hardware that you're running the big model on. There is a orchestration and resource competition problem inherent in that, and that is one of the constraints on speculation in general, is that draft tokens cost resources to create and cost software complexity to manage. And so if you have like infinitely recursive speculators, you add in quite a bit of that complexity on the actual implementation within the inference engine as well, not just in the training process.Vibhu [00:46:17]: I was gonna say, I would wonder if you could do similar, like distillation and pruning of, it's the same thing, it's just a model. Can we not just distill a lot of the weights, quantize the speculator, out of my domain? The question that also comes up is, this is all for big server workloads, right? How much of this applies to, say I have this MacBook, I wanna run Gemma really efficiently. Similar problems, not the same?Local AI vs. Data Center InferencePhilip [00:46:45]: Pretty different. I talked to Selo, about this on his podcast a couple weeks ago. The difference between inference engineering for the data center and for production workloads versus inference engineering for local AI, is that we start with fundamentally like different constraints and different goals. With local AI, it's how do I fit this model onto my hardware and then make it less dumb? And with data center influence, it's how do I load this model and then make it less slow? And we care about less dumb, and they care about less slow. But the local AI inference engineering ecosystem, I think has a lot for us to learn from in the data center space. They are experts in various forms of quantization, including dynamic quantization that we just don't touch, in the pruning, in the distillation, in the, layer removal. There'Ali [00:47:42]: Layer removal matters less.Philip [00:47:43]: Yeah. There'Ali [00:47:44]: No one loves pruning really.Philip [00:47:45]: Yeah. Well, but the, but they doVibhu [00:47:46]: Which is surprising, right? But that's, that's a whole different thingPhilip [00:47:48]: Just to fit something on the laptop.Ali [00:47:50]: Right.Philip [00:47:50]: So yeah, it's a, it's an interesting, it's an interesting space. Not necessarily that like their techniques make sense for us to do in the data center, because we have different resources and different goals, but more that the process as well as the openness of that field is something to, admire.Ali [00:48:12]: Yeah. Like to your point, like, certain optimizations that would. Like for instance, Turbo Quantum Sharper, like it made such huge hype on that and we did like a whole deep dive on Twitter and like said, what is it? How does it work? Why is it good or not? And it took off and it was implemented on local devices because your memory bandwidth is so slow on like a MacBook, for instance. But try putting the same thing on like an NVIDIA GPU on a B200 Turbo quant would not be. Like, it would not be used. Like, NVIDIA - Like, NVIDIA made it clear that this is not a good optimization, and we've seen it firsthand where the overhead of doing dequantization, quantization of, in the kernel itself with turbo quant kernel, each end is much slower than the time that you save from doing the bandwidth. ‘Cause on the B200s, you have like 3.5 terabytes per second. You don't need decrease the storage that much. You don't need to do, FP4 KV cache. You don't need to use a requant. There's, there's, there's better optimizations to be made. But on Edge devices, it's extremely important, it's extremely useful. So, seems to be, like, different optimizations there, but then they're all uniquely combined with like all you wanna quantize the model, you wanna do speculative decoding, like certain common prefixes with bothPhilip [00:49:18]: Principles.Ali [00:49:19]: Yeah, exactly. Exactly. Exactly.Philip [00:49:20]: They also do a lot of work on, model parallelism, especially over, heterogeneous topology, where you have, some sparks and they are wired together with, Ethernet, DGX sparks.Ali [00:49:35]: Yeah, this is the Exo Labs guys.Philip [00:49:36]: Yeah. You have, a nu

By Doug Green “AI goes blind at exactly the moment when you need it most.” In this Technology Reseller News podcast, Vishal Gupta, Director of Product Management at ZPE Systems, explains why AI-driven infrastructure management needs an independent path to the devices it is expected to monitor, troubleshoot and recover. AIOps platforms have become increasingly effective at detecting problems, correlating events and automating routine infrastructure operations. The problem, Gupta says, is that these systems often run on the same production infrastructure they manage. When a network outage or hardware failure occurs, the AI platform can lose both its connection to the affected equipment and access to the telemetry it needs to diagnose the problem. “That is the gap out-of-band fills,” says Gupta. Out-of-band management provides an independent management plane that remains separate from the production network. Even when the primary infrastructure is unavailable, IT teams—and increasingly AI agents—can still reach devices through console access, examine system logs and kernel messages, and take corrective action. Gupta compares the architecture to an airport. Aircraft use the runway for normal operations, while emergency and service vehicles have separate roads and infrastructure. If the runway becomes unavailable, the service infrastructure can still reach the aircraft. The same principle applies to resilient IT operations. An isolated management environment should have its own connectivity, security, routing, switching, storage and compute capabilities. It may also include failover connectivity through 4G, 5G or satellite services such as Starlink. Building Out-of-Band for a Larger Edge ZPE Systems developed its Nodegrid Net Services Router 2U, or NSR 2U, in response to customers operating increasingly large and complex edge environments. These environments can include branch offices, remote facilities, ships, oil rigs, cell sites and other locations outside traditional data centers. They frequently contain more devices, require greater bandwidth and have fewer trained personnel available on-site. The NSR 2U was designed around three priorities: greater capacity, increased resiliency and support for AI workloads. The modular platform offers 10 expansion-card slots, allowing customers to configure the system around their particular deployment. It also includes redundant, field-serviceable power supplies and fans, two NVMe storage slots with RAID support, four native 10-gigabit SFP+ ports and an increased Power over Ethernet budget. ZPE has even addressed the possibility that the out-of-band device itself could fail. Two NSR 2U systems can be interconnected so that one system can provide remote console, power and reset control for the other—effectively providing out-of-band management for the out-of-band infrastructure. Taking NVIDIA Jetson AI to Remote Locations ZPE Systems has also developed an NVIDIA Jetson AI Expansion Card for the Nodegrid NSR family. The card supports NVIDIA Jetson Orin Nano and Orin NX modules, providing local AI processing within the isolated management environment. This allows organizations to deploy AI agents close to the infrastructure and data they manage, without relying entirely on a remote cloud connection. A key capability is remote lifecycle management. IT teams can remotely flash the Jetson operating system, deploy or update AI agents and models, access the console, and power the device on, off or into recovery mode. Ordinarily, updating or recovering an edge AI device may require someone to travel to the location and connect directly to the hardware. ZPE's approach is intended to reduce those truck rolls while allowing organizations to manage distributed AI infrastructure centrally. Potential applications extend beyond AIOps. The platform can support real-time video analytics, object detection, smart recording, manufacturing quality control, sensor-data aggregation and local automation. GPIO and I2C interfaces also allow sensors measuring conditions such as temperature, vibration or voltage to feed information directly into locally running AI models. Asking the Hard Infrastructure Questions Gupta says much of the AI conversation remains focused on models, software and the token economy. Those areas are important, but they can obscure fundamental infrastructure questions. Where will an AIOps platform run? Can it survive the outage it is expected to resolve? Will it still have a path to the affected equipment? Can it access sufficiently accurate data to diagnose the problem and select the right recovery action? “If you can't answer these questions, then there's a gap in your AIOps strategy,” Gupta says. “No software and no model will fix it for you.” As AI becomes more autonomous, infrastructure resilience will determine whether AI agents can move beyond identifying failures to actually recovering from them. ZPE Systems is positioning isolated out-of-band infrastructure, the NSR 2U and edge-based Jetson AI processing as the foundation for making that transition possible. More at Enterprise Network Management Solution | ZPE Systems

Para precio y disponibilidad, vaya a este vínculo: https://amzn.to/4uGPMpu La estación de acoplamiento USB-C RayCue amplía las capacidades de una computadora portátil mediante un solo puerto USB-C. Permite conectar monitores externos a través de HDMI, dispositivos USB como teclados, ratones y discos duros, además de ofrecer conexión de red Ethernet, lectores de tarjetas SD y microSD, y carga de la computadora mediante USB-C Power Delivery. Es una solución práctica para crear una estación de trabajo más completa, reduciendo la cantidad de cables y facilitando la conexión de múltiples dispositivos al mismo tiempo.

Sun, 26 Jul 2026 15:00:00 GMT http://relay.fm/mpu/859 http://relay.fm/mpu/859 Tangential Gadgets 859 David Sparks and Stephen Robles David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. clean 5185 David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. This episode of Mac Power Users is sponsored by: Ecamm: Powerful live streaming platform for Mac. Decagon: The AI concierge for every customer. Get a personalized demo. Squarespace: Save 10% off your first purchase of a website or domain using code MPU. Links and Show Notes: Credits The Mac Power Users Stephen Robles David Sparks The Editor Jim Metzendorf The Fixer Kerry Provanzano More Power Users: Ad-free episodes with regular bonus segments Submit Feedback MPU on YouTube Tech Folio Laptop Backpack | USA Made | WaterField Designs Shinjuku Laptop Messenger Bag | USA Made | WaterField Designs Anker USB-C Hub Adapter (7-in-1) MOFT Dynamic Folio Case for iPad Pro 11-inch Anker USB-C Hub with Ethernet (8-in-1) Chipolo CARD Tracking Card Pro - Stellar Orange MOFT MagSafe Wallet Stand ESR Geo MagSafe Wallet Elevation Lab AirTag 10-Year Battery Compact Satechi FindAll Smart Glasses Case Anker Nano Charging Station ARK Uno Solid Wood MagSafe Stand Belkin MagSafe Charger 3-in-1 Charging Station KU XIU Qi2.2 Magnetic Wireless Charger Statik MagStack Pro USB-C Cable Anker 140W 4-Port GaN MacBook Charger Anker MagGo 3-in-1 MagSafe Charging Station $3,000 Kaleidescape Movie Box - YouTube UniFi Travel Router Jackery Solar Generator 300 Jackery SolarSaga 100W Solar Panel Trackables – Health Tracker App Outlet Shelf Wall Holder Denon AVR-S970H AV Receiver elago R5 Locator Case X2 Sofabaton Universal Remote Strato E Movie Player Lamicall MagSafe Tripod Stand Nillkin MagSafe Wall Mount Shower Phone Holder cobcobb Magnetic AirPods Strap UGREEN Magnetic Wireless USB-C to HDMI Adapter Here's the problem with Meta Glasses - YouTube Plaud.ai Spigen MagSafe Wallet Pouch Organizer TRMNL | ePaper dashboard to stay focused The Task Knife | Grovemade Leatherman ARC Multi-tool SnapFresh Cordless Electric Scissors Freewrite Signature Edition reMarkable Paper Tablet Why I'm (sort of) not worried about AI - YouTube Analogue 3D Roller Pro Carry-On | Peak Design

Sun, 26 Jul 2026 15:00:00 GMT http://relay.fm/mpu/859 http://relay.fm/mpu/859 David Sparks and Stephen Robles David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. clean 5185 David and Stephen turn the show on themselves: bags, chargers, MagSafe batteries, and fitness trackers. No tangential gadget is left behind. This episode of Mac Power Users is sponsored by: Ecamm: Powerful live streaming platform for Mac. Decagon: The AI concierge for every customer. Get a personalized demo. Squarespace: Save 10% off your first purchase of a website or domain using code MPU. Links and Show Notes: Credits The Mac Power Users Stephen Robles David Sparks The Editor Jim Metzendorf The Fixer Kerry Provanzano More Power Users: Ad-free episodes with regular bonus segments Submit Feedback MPU on YouTube Tech Folio Laptop Backpack | USA Made | WaterField Designs Shinjuku Laptop Messenger Bag | USA Made | WaterField Designs Anker USB-C Hub Adapter (7-in-1) MOFT Dynamic Folio Case for iPad Pro 11-inch Anker USB-C Hub with Ethernet (8-in-1) Chipolo CARD Tracking Card Pro - Stellar Orange MOFT MagSafe Wallet Stand ESR Geo MagSafe Wallet Elevation Lab AirTag 10-Year Battery Compact Satechi FindAll Smart Glasses Case Anker Nano Charging Station ARK Uno Solid Wood MagSafe Stand Belkin MagSafe Charger 3-in-1 Charging Station KU XIU Qi2.2 Magnetic Wireless Charger Statik MagStack Pro USB-C Cable Anker 140W 4-Port GaN MacBook Charger Anker MagGo 3-in-1 MagSafe Charging Station $3,000 Kaleidescape Movie Box - YouTube UniFi Travel Router Jackery Solar Generator 300 Jackery SolarSaga 100W Solar Panel Trackables – Health Tracker App Outlet Shelf Wall Holder Denon AVR-S970H AV Receiver elago R5 Locator Case X2 Sofabaton Universal Remote Strato E Movie Player Lamicall MagSafe Tripod Stand Nillkin MagSafe Wall Mount Shower Phone Holder cobcobb Magnetic AirPods Strap UGREEN Magnetic Wireless USB-C to HDMI Adapter Here's the problem with Meta Glasses - YouTube Plaud.ai Spigen MagSafe Wallet Pouch Organizer TRMNL | ePaper dashboard to stay focused The Task Knife | Grovemade Leatherman ARC Multi-tool SnapFresh Cordless Electric Scissors Freewrite Signature Edition reMarkable Paper Tablet Why I'm (sort of) not worried about AI - YouTube Analogue 3D Roller Pro Carry-On | Peak Design

Shortest Path Bridging (SPB) remains an intriguing and reliable fabric technology, offering a unique approach to Ethernet that differs from more widely deployed fabrics such as EVPN/VXLAN. Jeff Wilson joins Ethan and Drew to share his experience designing and managing SPB-based fabrics. Together they discuss how SPB simplifies multicast, essential hardware design considerations for scaling,... Read more »

Shortest Path Bridging (SPB) remains an intriguing and reliable fabric technology, offering a unique approach to Ethernet that differs from more widely deployed fabrics such as EVPN/VXLAN. Jeff Wilson joins Ethan and Drew to share his experience designing and managing SPB-based fabrics. Together they discuss how SPB simplifies multicast, essential hardware design considerations for scaling,... Read more »

Shortest Path Bridging (SPB) remains an intriguing and reliable fabric technology, offering a unique approach to Ethernet that differs from more widely deployed fabrics such as EVPN/VXLAN. Jeff Wilson joins Ethan and Drew to share his experience designing and managing SPB-based fabrics. Together they discuss how SPB simplifies multicast, essential hardware design considerations for scaling,... Read more »

Thu, 23 Jul 2026 19:15:00 GMT http://relay.fm/connected/613 http://relay.fm/connected/613 Long-Distance Ethernet Cable 613 Federico Viticci, Stephen Hackett, and Myke Hurley This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. clean 4352 This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. This episode of Connected is sponsored by: Fitbod: Get stronger, faster with a fitness plan that fits you. Get 25% off your membership. Backblaze: Unlimited, easy data protection. Try it for free today and get 20% off with code connected20. Links and Show Notes: Get 15% off a new annual Connected Pro subscription (or reactivate an expired one!) with coupon code tinyheads15 between now and July 31. Get Connected Pro: Preshow, postshow, no ads. Submit Feedback TestFlight: One Step Forward, Two Steps Back - MacStories App Store Chaos - MacStories Open Minis Is the iOS Agent I Wish Siri AI Could Be - MacStories Open Minis — Your Private On-Device AI Agent Samsung Z Fold 8 (Wide) Impressions: Better Than I Thought! - MKBHD - YouTube Samsung's Galaxy Z Fold 8 is a fun new shape for foldables | The Verge The Short Samsung Fo

Thu, 23 Jul 2026 19:15:00 GMT http://relay.fm/connected/613 http://relay.fm/connected/613 Federico Viticci, Stephen Hackett, and Myke Hurley This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. clean 4352 This week: App Store madness and TestFlight chaos, Federico has a new favorite app, and Myke is bringing news about a potential precursor to the folding iPhone. This episode of Connected is sponsored by: Fitbod: Get stronger, faster with a fitness plan that fits you. Get 25% off your membership. Backblaze: Unlimited, easy data protection. Try it for free today and get 20% off with code connected20. Links and Show Notes: Get 15% off a new annual Connected Pro subscription (or reactivate an expired one!) with coupon code tinyheads15 between now and July 31. Get Connected Pro: Preshow, postshow, no ads. Submit Feedback TestFlight: One Step Forward, Two Steps Back - MacStories App Store Chaos - MacStories Open Minis Is the iOS Agent I Wish Siri AI Could Be - MacStories Open Minis — Your Private On-Device AI Agent Samsung Z Fold 8 (Wide) Impressions: Better Than I Thought! - MKBHD - YouTube Samsung's Galaxy Z Fold 8 is a fun new shape for foldables | The Verge The Short

Michael and Jake catch up ahead of Laracon in Boston, beginning with Michael's increasingly complicated home extension. Jake shares a family trip to Cedar Point, then the conversation moves to how a Laravel Cloud weekend challenge inspired Michael to build Fitrack, a personal app for coordinating training around a single fitness goal. We then move on to home network and security upgrades, including a larger rack cabinet, a new NAS, locally accessible alarms and CCTV, Power over Ethernet cameras, Home Assistant, HomeKit, RTSP feeds, and using Claude to handle unfamiliar smart-home configuration. We also discuss Kitze hacking his NordicTrack treadmill and Michael's experience connecting his own treadmill with Apple Fitness.Finally, Michael and Jake look ahead to Laracon, discuss conference merchandise and accommodation in Boston, and finish with a conversation about donuts, candy, and the lengths parents go to when quietly disposing of their children's leftover sweets.Show linksCedar PointLaravel CloudTaylor's Laravel Cloud weekend shipping challengeEdwin's Google Meet alternativeFitrackDavid HemphillHome AssistantAqara camerasKitze's treadmill hackCaleb Porzio on code imprinting on our brainsLaracon USMoxy Boston Downtown

Send us Fan MailTwo competitors deciding to stop fighting and build together always raises one question: what changes for the customers the next morning. We start with the BT and Verizon International joint venture and dig into what a shared global network footprint could mean across SD-WAN, MPLS, Ethernet and cloud connectivity, including the uncomfortable parts like orchestration changes, duplicated POPs and the possibility of forced migrations. If you run global sites, this is the kind of deal that can quietly reshape your WAN roadmap.Then we shift to the money and physics behind the cloud: Virginia's new per kilowatt-hour data center electricity tax. It sounds small until you apply it to hyperscale consumption, and the inclusion of self-generated power closes an obvious workaround. We talk through why Northern Virginia's data center corridor matters, how policy spreads state to state, and why the “they'll just pass the cost on” argument is less theory than a pricing strategy you eventually see in your cloud bill.From there, it's security and governance: the White House post-quantum cryptography executive order and the accelerating reality of quantum risk, including “harvest now, decrypt later.” We connect PQC deadlines to refresh cycles, vendor readiness, and the practical work of migrating both infrastructure and applications. We also hit identity head-on with Cisco's plan to bring identity lifecycle security into Splunk's agentic SOC, because non-human identities and AI agents are changing what least privilege even means.We close with the growing pattern of government intervention in frontier AI models and what it does to businesses building on specific model capabilities. If you found this useful, subscribe, share the show with a friend and leave a review so more people can find it.Check out the Monthly Cloud Networking Newshttps://docs.google.com/document/d/1fkBWCGwXDUX9OfZ9_MvSVup8tJJzJeqrauaE6VPT2b0/Visit our website and subscribe: https://www.cables2clouds.com/Follow us on BlueSky: https://bsky.app/profile/cables2clouds.comFollow us on YouTube: https://www.youtube.com/@cables2clouds/Follow us on TikTok: https://www.tiktok.com/@cables2cloudsMerch Store: https://store.cables2clouds.com/Join the Discord Study group: https://artofneteng.com/iaatj

Sporlan S3C, Comm Loops, Fireman's Dump Line, Brett's Retiring, and Noodling Episode 528Fishing Fails, Target Startup Nightmares & Chicago Fireman's Dump Lines | Advanced Refrigeration PodcastBrett Wetzel and Kevin Compass open with fishing stories and then jump into a rough Target CO2 startup plagued by miswired comm loops, mixed case-control hardware, and constant walking in a multi-level downtown layout. They discuss troubleshooting methods for crisscrossed MODbus wiring, common causes of shorts and opens, and an electrician mistake that blew an XM controller apart. The hosts explain Chicago's “fireman's dump line” code requirement and how piping it off the bottom of a suction header can trap oil and cause recurring low-oil issues. They also flag a backwards high-pressure control/relay setup on a Target backup CO2 unit, share practical tips for S3C/Corelink/CC200 case controllers, Ethernet jumpers, termination, protecting solid-state outputs, transducer orientation, and water-damaged controllers from case washing.

Fishing Fails, Target Startup Nightmares & Chicago Fireman's Dump Lines | Advanced Refrigeration PodcastBrett Wetzel and Kevin Compass open with fishing stories and then jump into a rough Target CO2 startup plagued by miswired comm loops, mixed case-control hardware, and constant walking in a multi-level downtown layout. They discuss troubleshooting methods for crisscrossed MODbus wiring, common causes of shorts and opens, and an electrician mistake that blew an XM controller apart. The hosts explain Chicago's “fireman's dump line” code requirement and how piping it off the bottom of a suction header can trap oil and cause recurring low-oil issues. They also flag a backwards high-pressure control/relay setup on a Target backup CO2 unit, share practical tips for S3C/Corelink/CC200 case controllers, Ethernet jumpers, termination, protecting solid-state outputs, transducer orientation, and water-damaged controllers from case washing.

Si parla di batteria e funzioni bloccate di iOS, lettori RSS e piccole chicche da Mac, LackRack e video della settimana, workflow tra YouTube e Todoist, Octopus Energy in Home Assistant, irrigazione smart, screenshot lunghi e limiti dello sviluppo su iOS.

This show has been flagged as Clean by the host. This series is dedicated to exploring little-known—and occasionally useful—trinkets lurking in the dusty corners of UNIX-like operating systems. In UNIX Curio #8 ( HPR episode 4657 ), I talked about using standard utilities to compare files. Left unmentioned, however, was a method commonly used today—the hash function. As I've stated in previous entries, while I am an engineer, I don't have a background in computer science, so my understanding of the mathematics is limited. But I can give a practical description of what a hash function does. It takes an input, performs a set of calculations on it, and produces an output. As hash functions are practically used, the input is a set of bytes, such as a file or another piece of data like a password. The output is a numerical value in a fixed range—most often, expressed as hexadecimal characters. Because this "hash value" can always be represented in a certain number of bytes, its length as printed is usually a constant number of characters, padded with leading zeros if necessary. This episode will not cover the use of hashes in programming, focusing instead on using them to validate data. A hash function, or more specifically, a cryptographic hash function, has an additional property. It should be very difficult to predict what changes to the input would be required to produce a specific change in the output. An older, related concept is called a "checksum". While these are designed to vary when the input data is damaged or digits are transposed, they do not necessarily have that last property mentioned for cryptographic hashes. You have probably already encountered a checksum, even if you didn't recognize it. On a 16-digit number assigned to a Mastercard or Visa 1 credit or debit card, the first six digits identify the card issuer (such as a bank), the next nine digits are assigned to you by the issuer, and the last digit is a check digit. The check digit is calculated using the values of the previous 15 digits, and it is a simple way to avoid typos in entering a card number. In another example, every Ethernet frame that your devices send or receive includes a checksum 2 to help ensure that the contents weren't scrambled in transit. This is 32 bits long and is called a cyclical redundancy check, commonly referred to as a CRC. A CRC is also used in many other places—for example, the .zip file format includes one for each archive member, and this allows a program extracting files from the archive to identify if any were damaged. Our UNIX Curio for today is another example, the cksum utility 3 . It generates a 32-bit CRC based on the Ethernet algorithm. It operates on either a named file or standard input and outputs the CRC value, the length of the input, and the pathname if a file was given as an argument. Unlike most modern hashing programs, the checksum is printed as a decimal integer and is not padded, so it can be anywhere from one to ten digits long. The length value is the number of bytes in the input (actually specified as the number of octets , as systems could potentially use a byte that isn't eight bits long), also expressed as a decimal integer. There are two major ways that one could use cksum to check the validity of a file. First, if you are transferring a file from one UNIX-like system to another, you could run cksum against it on both systems and check that the CRC and length are the same. The utility can also be given multiple filenames as arguments, which would generate a list that can then be compared. The second way would be for someone publishing a file or set of files to also publish the CRC values, lengths, and names so that people downloading them could verify that they match. However, I don't think the practice of publishing lists like this really started until more recent hash functions like MD5 and SHA-1 came about so it is unlikely that anyone would publish CRC values instead. The advantage of these tools should be pretty obvious in comparison to cmp , one of the utilities discussed in UNIX Curio #8. To verify a file using cmp , you need two files to compare—if you're trying to check a large file you downloaded, you would need to spend the time and bandwidth to download a second copy. And if they didn't match, you would have no idea which of the two, if either, was correct. By contrast, cksum is quicker to run, doesn't require downloading a massive amount of excess data, and if run against the original file, makes clear what the correct value is. This utility is a follow-on from a program called sum , which operated very much the same. I had a bit of trouble tracking down the exact development history, but what seems clear is that two different variants 4 were popular: a BSD version and a System V version. Both output 16-bit checksums, but used different algorithms so they didn't give the same results. Also, the BSD version printed the length of the input data as the number of 1,024-byte blocks, while the System V version instead gave a count of 512-byte blocks. (Some sources claim that System V sum generates a 32-bit checksum 5 , which could possibly be true internal to the algorithm, but I have tested several independent implementations of the utility and all of them output a 16-bit value for both the System V and BSD algorithms.) From what I can tell, the BSD version 6,7 came first; it was in 3BSD but probably appeared even earlier. An identical copy of BSD's sum was included with UNIX/32V 8,9 , which was AT&T's 1979 port of Seventh Edition UNIX to the VAX and became one of the ancestors of System III. The divergence seems to have started with System III, released in 1980; its version of the sum utility 10,11 changed to a new default algorithm, though it could be made to use the BSD algorithm via the -r option. System V looks to have kept the same behavior as System III. It's not clear to me why this algorithm is universally called the "System V algorithm" rather than the "System III algorithm"; perhaps it is because System V saw much more widespread use. Instead of trying to reconcile these differences, the POSIX committee decided to create a new utility with a unique name, use a separate algorithm entirely, and avoid the block-length dispute by printing the length in octets instead of blocks. I should point out that POSIX states that the CRC algorithm for cksum does not strictly meet the mathematical definition of a "checksum". I don't know enough to say exactly why it doesn't qualify or to say whether either of the sum algorithms do. However, in less-formal usage the term "checksum" has gathered the meaning of any value used to represent or validate a set of data, so I am fine with using it no matter the technical details of the algorithm. When two different inputs produce the same checksum or hash value, this is called a "collision". Because the output value has a limited range, there are an infinite number of possible inputs that could produce a collision. From a practical standpoint the possibilities are more limited—the majority of these inputs are larger than the number of atoms in the universe, which can't fit on any machine. Unlike a cryptographic hash algorithm, the CRC is not specifically designed to resist an attacker crafting a malicious input that would cause a collision. However, it should be sufficient to detect accidental damage. Programs implementing more modern cryptographic hash algorithms are superior to the checksum utilities in avoiding collisions (whether malicious or accidental), but there are still three advantages that the older programs have. First, a system running a historical operating system might not have the hash programs available, but is more likely to have cksum or sum already included. Second, the checksum values are much shorter than the hashes output by the newer programs, making them easier for a user to compare by looking at them. This advantage is not as great as it might appear at first, because a common way to check a hash these days is to save a list of hashes and filenames—the hash programs can use that and do the comparison themselves, sparing the user from having to validate it character by character. The third advantage is that cksum prints the input length in bytes. This greatly limits the number of inputs that could be maliciously crafted to create a collision. I did a moderate amount of research on implementations of modern cryptographic hash algorithms and found that some, such as MD5, SHA-1, and SHA-2, do use the length of the input (often termed "message length" in the literature) as part of the material fed in to the algorithm, but none of the hashing utilities present this length to the user as part of its output. There are two possible reasons for this that seem evident to me. First, if one is hashing a password, you would certainly not want to give a clear indication of its length—that would give any attacker a massive head start on guessing the password. However, that doesn't explain why one would avoid printing the input length for a file that is made publicly available. Second, it is convenient in many contexts, such as database entries or in software (such as git ), for the hash to be a fixed length. Including an extra value that can be of variable length would complicate those use cases. However, the length value could simply be dropped and they would be no worse off than they are currently. Historically on UNIX, password hashing was treated differently from checksumming files— the crypt() function 12 was used for passwords while sum and later cksum were used to confirm a file's integrity. So even rather early on, these two use cases employed algorithms with different properties, but I haven't dived into the history deeply enough to know how intentional this was. My discussion in this episode focuses on the file use case, so understand that I'm largely avoiding the topic of password hashing. Digital signatures are yet another use case, one that I'm ignoring entirely. Every few years, some security researcher declares a particular hash algorithm to be "broken" and that everyone should move over to a new one, which generally has a longer hash. While the larger hash space certainly reduces the opportunity for collisions, this disrupts workflows, such as publishing information about software releases by e-mail, which still tends to observe a 78-character limit on each line 13 , making it harder to include a list of hashes with filenames next to them. This is in addition to the work of modifying software and scripts to use the new algorithm and managing how to treat past data. It seems to me that publishing the input length along with the hash would make it far more difficult to craft a malicious input that matches both, but I haven't found discussion of that during my investigation. (See the Appendix for a possible implementation.) Perhaps someone listening can record a response episode for HPR explaining that. References: Payment card number https://en.wikipedia.org/wiki/Payment_card_number Ethernet frame: Frame check sequence https://en.wikipedia.org/wiki/Ethernet_frame#Frame_check_sequence Cksum specification https://pubs.opengroup.org/onlinepubs/009695399/utilities/cksum.html GNU coreutils manual: sum https://www.gnu.org/software/coreutils/manual/html_node/sum-invocation.html FreeBSD 15.0 sum manual page https://man.freebsd.org/cgi/man.cgi?query=sum&sektion=1&manpath=FreeBSD+15.0-RELEASE+and+Ports 3BSD sum manual page https://www.tuhs.org/cgi-bin/utree.pl?file=3BSD/usr/man/man1/sum.1 3BSD sum source https://www.tuhs.org/cgi-bin/utree.pl?file=3BSD/usr/src/cmd/sum.c UNIX/32V sum manual page https://www.tuhs.org/cgi-bin/utree.pl?file=32V/usr/man/man1/sum.1 UNIX/32V sum source https://www.tuhs.org/cgi-bin/utree.pl?file=32V/usr/src/cmd/sum.c System III sum manual page https://www.tuhs.org/cgi-bin/utree.pl?file=SysIII/usr/src/man/man1/sum.1 System III sum source https://www.tuhs.org/cgi-bin/utree.pl?file=SysIII/usr/src/cmd/sum.c Crypt specification https://pubs.opengroup.org/onlinepubs/009695399/functions/crypt.html RFC 2822: Internet Message Format: Line Length Limits https://datatracker.ietf.org/doc/html/rfc2822#section-2.1.1 OpenSSH 10.1 released https://lwn.net/ml/all/dd12623ae86aa5eb@cvs.openbsd.org/ Appendix The MD5 hash algorithm was (and still is) widely used, but many people characterize it as being "broken" and discourage its use. Let us imagine a variant of this, called MD5.L, where the normal MD5 hash is followed by a "." character and the input length expressed as a hexadecimal number. Take, for example, the e-mail message announcing the release of OpenSSH 10.1 14 . At the bottom, it includes an SHA-1 hash and an SHA-2 256-bit hash for the available gzipped tar files. That longer hash is encoded with Base64 because if it were given as a hexadecimal number, it would make the line longer than 78 bytes. The MD5.L hash of the file would be one character shorter than the SHA-1 hash, as shown below. (The extra length of the name makes them both consume the same number of characters. The hashes shown are for the "portable" version of OpenSSH.) Some people claim SHA-1 is also broken, seeking to have people use newer and longer hash functions. For an attacker to compromise MD5.L in this example, they would not only have to create a valid tar file compressed with gzip containing a malicious payload having the right MD5 hash, that file would have to be exactly 1,972,831 bytes long (the decimal equivalent of 1e1a5f). While there are still many possible inputs that could be tried (256 1972831 , to be exact*), this is far fewer than the infinite possibilities for plain MD5, SHA-1, or SHA-2. If for some reason it is super important to have a fixed hash length, let's imagine another variation called MD5+L. In this one, instead of L being the input length, it is the input length modulo one terabyte (2 40 bytes), which can be represented by 10 hexadecimal characters, left-padded with zeros. While this approach substantially increases the number of possible inputs an attacker could try, it is likely that an intended victim would notice that the file they downloaded is larger (or smaller) than expected by that much. The MD5+L hash is longer than a SHA-1 hash, but still shorter than a 256-bit SHA-2 hash. SHA1 (openssh-10.1p1.tar.gz) = 7fd17b99d1beffb47cd380d64079e920bb0bd91f SHA256 (openssh-10.1p1.tar.gz) = ufx6K4JXlGem8vQ+SoHI4d/aYU3bT5slWq/XAgu/B1g= MD5.L (openssh-10.1p1.tar.gz) = 80dd9bb00a86519934710d05903fdf07.1e1a5f MD5+L (openssh-10.1p1.tar.gz) = 80dd9bb00a86519934710d05903fdf07+00001e1a5f Of course, if MD5 is considered to be too weak even with the inclusion of the length, one could produce a ".L" or "+L" version of any hash function. However, longer hashes will end up running into the 78-character limit. *This is a number with 4.75 million digits that the bc utility on my laptop took almost 5 minutes to calculate. Provide feedback on this episode.

Big thank you to DeleteMe for sponsoring this video. Use my link https://joindeleteme.com/Bombal to receive a 20% discount or use the QR Code in the video. Also, a Big thanks to Cisco for sponsoring my trip to Cisco Live Vegas 2026. Do you really need Cat 8 Ethernet for your home network in 2026? And what actually happens when you touch the end of a fiber optic cable? We break it all down with hands-on demos. At Cisco Live, I caught up with Jim from Fluke Networks to dive deep into the physical layer of networking. We strip away the marketing hype and look at the real-world performance of both copper and glass. In this video, we cover everything from the crucial safety reasons why you should never look into a fiber cable, to a microscope demonstration showing how a single speck of dirt or finger oil can cause frame errors and take down your SFP modules. We also run a live cable certification test on Cat 6 vs Cat 8 to show you exactly why Cat 6A is all you actually need for your home network, and how pros use an OTDR to find exact breaks in long fiber runs. If you want to build or troubleshoot networks effectively, you need to understand the physical layer. // Jim Davis from Fluke Networks SOCIAL // LinkedIn: / jimdavisflukenetworks Website: https://www.flukenetworks.com/ // David's SOCIAL // Discord: discord.com/invite/usKSyzb Twitter: www.twitter.com/davidbombal Instagram: www.instagram.com/davidbombal LinkedIn: www.linkedin.com/in/davidbombal Facebook: www.facebook.com/davidbombal.co TikTok: tiktok.com/@davidbombal YouTube: / @davidbombal Spotify: open.spotify.com/show/3f6k6gE... SoundCloud: / davidbombal Apple Podcast: podcasts.apple.com/us/podcast... // MY STUFF // https://www.amazon.com/shop/davidbombal // SPONSORS // Interested in sponsoring my videos? Reach out to my team here: sponsors@davidbombal.com // MENU // 0:00 - Coming up 0:31 - Introduction 1:31 - Don't touch fiber optic cables! 04:53 - Single vs Multi-mode fiber 08:04 - Don't look directly into a fiber optic cable! 09:13 - DeleteMe sponsor segment 11:03 - How to find a break in a cable 16:01 - Can ethernet copper cables get dirty? 18:12 - How to test ethernet cables 20:00 - The future of ethernet 21:45 - Conclusion Please note that links listed may be affiliate links and provide me with a small percentage/kickback should you use them to purchase any of the items listed or recommended. Thank you for supporting me and this channel! Disclaimer: This video is for educational purposes only. #fluke #fiberoptics #ethernetcable

Copper twisted pair cabling serves as a fundamental component of Ethernet infrastructure and Ethan and Holly are here to break down how it works. They discuss the technical differences between cabling categories, how wire twisting cancels out electromagnetic interference, and share practical guidance on installation standards and testing methodologies. Episode Links: Watch this episode on... Read more »

Copper twisted pair cabling serves as a fundamental component of Ethernet infrastructure and Ethan and Holly are here to break down how it works. They discuss the technical differences between cabling categories, how wire twisting cancels out electromagnetic interference, and share practical guidance on installation standards and testing methodologies. Episode Links: Watch this episode on... Read more »

Big thanks to NTT DATA for sponsoring this video and to Cisco for sponsoring my trip to Cisco Live Vegas. For more information about NTT DATA and Cisco's partnership, click here https://services.global.ntt/en-us/abo... 1000W touch safe power sounds impossible, but Class 4 Fault Managed Power is changing how we think about power delivery for AI data centers, buildings, networking, and IT infrastructure. In this interview, we look at how Cisco, NTT, and Panduit are approaching the next generation of energy networking systems. Traditional Power over Ethernet is useful for endpoints like access points, cameras, sensors, and phones, but PoE tops out around 100 watts. Fault Managed Power introduces a new class of power delivery that can deliver much higher wattage while intelligently shutting down when it detects a fault, short circuit, or unsafe condition. The conversation explains why this matters now. AI is driving massive demand for data center compute, but that compute requires more power, more cooling, and more efficient infrastructure. The guests explain how AC-to-DC conversion losses add up across multiple power conversion stages, how transmission and distribution losses waste energy before it even reaches the building, and why many legacy data centers spend a large amount of energy on cooling and non-useful power. A key example from the discussion: an average 5 megawatt, air-cooled data center moving to direct liquid cooling and fault managed power could save 75% of its energy cost. That same energy profile could also translate into 2.3x the compute for the same amount of energy. We also discuss touch-safe DC power up to 400V, why DC-to-DC power matters for data centers, how this technology could reduce material and labor complexity, and why network engineers, electricians, data center operators, and IT professionals should understand Class 4 Fault Managed Power. This is not just about sustainability. It is about energy cost, AI infrastructure, safer power delivery, faster deployment, data center efficiency, retrofits, new builds, and the future of networking. // Stephen Kelly SOCIAL // LinkedIn: / stephen-kelly-70737a9 // Denise Lee SOCIAL // LinkedIn: / deniseleeyeh // Website REFERENCE // https://www.nttdata.com/global/en/ // David's SOCIAL // Discord: discord.com/invite/usKSyzb Twitter: www.twitter.com/davidbombal Instagram: www.instagram.com/davidbombal LinkedIn: www.linkedin.com/in/davidbombal Facebook: www.facebook.com/davidbombal.co TikTok: tiktok.com/@davidbombal YouTube: / @davidbombal Spotify: open.spotify.com/show/3f6k6gE... SoundCloud: / davidbombal Apple Podcast: podcasts.apple.com/us/podcast... // MY STUFF // https://www.amazon.com/shop/davidbombal // SPONSORS // Interested in sponsoring my videos? Reach out to my team here: sponsors@davidbombal.com // MENU // 0:00 - Coming up 0:53 - New cables 02:22 - Fault-Managed Power System explained 03:18 - Cisco & NTT DATA partnership 04:39 - Advantages of the new cables 08:57 - Sustainability 11:26 - AC vs DC 13:14 - Easy to install // Free training 14:50 - How new cables fit into old buildings 17:44 - Embrace the new 19:35 - Will electricians be replaced? 20:49 - Summary 21:50 - Free training and learning 23:09 - Stay connected // Conclusion Please note that links listed may be affiliate links and provide me with a small percentage/kickback should you use them to purchase any of the items listed or recommended. Thank you for supporting me and this channel! Disclaimer: This video is for educational purposes only. #poe #class4 #ntt

Make a Logo on Fiverr The Nearity 360 Alien is not your typical conference camera. Instead of sitting at the front of the room like a traditional webcam or PTZ camera, this tall, table-mounted system is designed to sit in the middle of the action and capture nearly everyone around it. With four built-in cameras, AI framing, multiple meeting modes and optional wireless connectivity, the Nearity 360 Alien is built for boardrooms, hybrid meetings, presentations and conference setups where one camera needs to do the work of several. A Conference Camera Built for the Middle of the Table The Nearity 360 Alien is a 4K wireless camera system designed to capture a room from the center of the table. It uses four cameras to create a panoramic meeting view, then relies on AI to identify speakers and keep people framed. That makes it especially useful for rooms where people are seated around a table instead of lined up in front of a screen. Rather than forcing everyone to crowd around a laptop webcam, the Nearity 360 Alien gives remote participants a clearer view of who is speaking and where they are in the room. The Design Is Big, But That Works in Its Favor The first thing you notice about the Nearity 360 Alien is its size. This is not a tiny puck camera. It is tall, noticeable and built to sit above the table surface. That height is actually a benefit. A smaller conference camera can end up shooting people from too low of an angle, especially in a boardroom. The taller body helps the camera capture faces more naturally, and the built-in quarter-inch mount means you can also place it on a tripod if you need a better eye-level shot. There is one physical limitation: despite the 360 branding, there is a small dead spot where the cameras do not overlap. In most rooms, that will not be a major issue if you position that side toward a wall, monitor or unused side of the table. What Comes in the Box Inside the box, the Nearity 360 Alien includes the main camera unit, power adapter, remote control, USB-A to USB-C cable and a wired microphone puck with a mute button. The microphone connects directly to the camera, and a second mic can be added for larger table setups. The remote is one of the more useful accessories. It lets you change views, adjust modes, mute audio, mute video and control settings without needing to sit next to the computer. That is a big plus if the camera is being used in a meeting room where the laptop or production system is not within easy reach. Ports and Connectivity The Nearity 360 Alien includes DC power, USB-C, Ethernet, HDMI and microphone ports. The HDMI port is full-size, which is a welcome detail because it avoids the need for micro-HDMI or mini-HDMI adapters. USB-C can connect the camera to a computer for Zoom, Teams, Google Meet, OBS, vMix, Wirecast or other video software. HDMI can send video directly to a recorder, switcher or display setup. Ethernet can be used for network configuration, while the optional NA20 wireless dongle allows the camera to connect wirelessly to a computer. That wireless option is one of the stronger parts of the system. Once paired, the NA20 shows up as a wireless video and audio source, letting you use the Nearity 360 Alien in meeting apps or production software without running a USB cable across the table. Meeting Modes: Discussion, Presentation and Global The Nearity 360 Alien includes three main modes: Discussion Mode, Presentation Mode and Global Mode. Discussion Mode Discussion Mode is the best choice for meetings with multiple people talking around the table. The camera identifies faces and can show multiple participants in framed sections, while still keeping the panoramic view visible. In testing, the system was able to recognize more than one person in the room and place them into the split layout. When someone speaks, the system can highlight or shift attention toward that person, helping remote attendees follow the conversation. Presentation Mode Presentation Mode is meant for a meeting where one person is leading the discussion. This mode keeps the presenter as the focus while still letting the room remain visible. For lectures, board updates, demos or hybrid presentations, this is probably the mode most people will use. Global Mode Global Mode gives a broader room view. It is useful when you want everyone visible and do not need the AI to focus tightly on individual speakers. Video Quality and AI Framing The Nearity 360 Alien uses a Starvis CMOS sensor and offers USB 4K output, with HDMI output up to 1080p. The camera also includes image controls such as brightness, saturation, contrast, hue, white balance, HDR and regional frequency settings. In a well-lit room, the image looks clean and usable for meetings. Studio lighting looked especially good, and the camera handled face framing well. In a real meeting environment, it did a solid job following the conversation and keeping speakers visible. Lighting matters, though. Like most conference cameras, the Nearity 360 Alien performs best when the room is evenly lit. In darker or more complicated environments, such as a stage or band-meeting setup with speakers and instruments in the background, the AI can occasionally get confused. Audio: Built-In Mics, Speaker and External Mic Pucks The camera has built-in microphones on top, along with a speaker near the bottom. The included external microphone puck adds more flexibility because it can be placed closer to the person speaking or moved toward the center of the conversation. There is also a mute button on the mic puck, which is useful in a meeting where people need quick local control. Adding a second mic gives the system better coverage on both sides of a larger table. One practical tip: place the camera or mic puck on a soft surface like a mouse pad if the table picks up vibration. That can help reduce thumps, glass clinks and table noise before the audio processing has to clean it up. Software and Controls The Nearity software gives access to device settings, firmware updates, image controls, mode selection, network settings and options like flipping the image or enabling a watermark. The app is not the only way to use the camera. One of the best parts of the Nearity 360 Alien is that it can work directly over HDMI or USB without making the setup feel overly complicated. The remote also gives quick access to many of the key controls. Wireless Setup With the NA20 Dongle The NA20 wireless dongle is what turns the Nearity 360 Alien into a more flexible 4K wireless camera solution. After plugging the dongle into a computer and pairing it with the camera, it appears as a camera and audio source in meeting software. That means you can bring it into Zoom, Teams, Google Meet, OBS, vMix or other tools as a video source. Once it is working wirelessly, the only cable that still needs to remain attached to the camera is power. For conference rooms where cable clutter is a problem, that is a major win. Pros The Nearity 360 Alien does a strong job replacing a small multi-camera meeting setup with one device. It captures the room, identifies speakers, offers several useful layouts and works with common meeting and production apps. The HDMI output is convenient, the remote control is genuinely useful, the mic puck adds flexibility and the wireless dongle makes the system easier to deploy in rooms where you do not want cables running across the table. It also works well for hybrid meetings where people in the room and people online both need to feel included. Cons The main downside is the small dead spot behind the camera. It is not a true edge-to-edge 360-degree view, so placement matters. The camera is also fairly large, which may not fit every table setup. And while the AI tracking works well in normal meeting conditions, more chaotic environments with multiple visual distractions can make it less reliable. Is the Nearity 360 Alien the Best Wireless Conference Camera? The Nearity 360 Alien makes a strong case for itself if you need a conference camera that can handle real meeting rooms, not just one person sitting in front of a laptop. It is especially useful for boardrooms, team meetings, hybrid presentations, nonprofit meetings, training rooms and any space where multiple people need to be seen and heard clearly. The combination of 4K USB output, HDMI, AI framing, external mic support, meeting modes and optional wireless connectivity makes it more flexible than a standard webcam. It is not perfect, and the small blind spot means you need to think about placement. But for a room where one conference camera needs to cover almost everyone, the Nearity 360 Alien is an impressive option. Get this Conference Camera here! Check out the Geekazine Merch, including "I AM AI " T-Shirt.  Thanks for reading! Don't forget to subscribe to Geekazine: RSS Feed - YouTubeTwitter - Facebook Tip Me via Paypal.me Send a Tip via Venmo RSS Bandwidth by Cachefly Get a 14 Day Trial Be a Patreon: Part of the Sconnie Geek Nation! Reviews: Geekazine gets products in to review. Opinions are of Geekazine.com. Sponsored content will be labeled as such. Read all policies on the Geekazine review page.  Reviews: Geekazine is also an affiliate of Amazon Last Updated on June 25, 2026 5:01 pm by Jeffrey PowersThe post Nearity 360 Alien Review: Best Wireless Conference Camera? appeared first on Geekazine.

MACsec (IEEE 802.1AE) encrypts Ethernet frames hop-by-hop at Layer 2 — before traffic even hits IP — making it one of the strongest protections you can put on wire. It’s been in the standards for years, hardware support is widespread, and yet most organizations aren’t running it. JJ and Drew dig into why: the hardware... Read more »

MACsec (IEEE 802.1AE) encrypts Ethernet frames hop-by-hop at Layer 2 — before traffic even hits IP — making it one of the strongest protections you can put on wire. It’s been in the standards for years, hardware support is widespread, and yet most organizations aren’t running it. JJ and Drew dig into why: the hardware... Read more »

Send us Fan MailInfoComm can sound like “an AV show,” but if you pull cable for a living, it's really a massive showcase of where low voltage work is headed next. We sit down with Bob Neyens, VP at Vertical Cable and a long-time InfoComm regular, to translate the pro AV world into plain English and show exactly why structured cabling installers, project foremen, project managers, and designers can get real value from attending.We talk about what you'll actually find at InfoComm: professional audio video for commercial spaces like boardrooms, houses of worship, bars, stadiums, and even hospitals, with tons of crossover into IP networks and familiar infrastructure like Cat 6, Cat 6A, fiber, racks, and connectors. Bob shares practical exhibit hall tactics for first-timers, including how to avoid the most chaotic day, when to schedule booth conversations, and how manufacturer events and networking work after hours.Then we get specific about skills and career growth: AVIXA training, CTS certifications like CTS-I and CTS-D, and why the show floor can teach you what a classroom can't by letting you see complete systems running together. We also dig into tech that matters for 2026, including the move from Cat 6 to Cat 6A for higher bandwidth and distance, plus Power over Ethernet and why copper quality matters as current loads increase. If you want a new revenue stream that stays close to your current skill set, this is the roadmap.Subscribe for more conversations that connect the low voltage community, and if this helps, share it with a tech who's been curious about pro AV and leave us a rating or review.Support the showKnowledge is power!  Make sure to stop by the webpage to buy me a cup of coffee or support the show at https://linktr.ee/letstalkcabling .  Also if you would like to be a guest on the show or have a topic for discussion send me an email at chuck@letstalkcabling.com Chuck Bowser RCDD TECH#CBRCDD #RCDD

Send us Fan MailWe catch up on what we've been doing lately, from gardening plans and convention runs to building skills for cybersecurity and digital forensics. Then we go full hobby mode as we debate music takes, break down the biggest gaming showcases, and vent about why some studios keep missing the moment. • figuring out where to plant sun-heavy flowers and bonsai seeds without squirrels ruining them • quick MomoCon stop that still turns into merch, shirts, and glassware • starting an ethical hacking class and aiming long-term at digital forensics • upgrading to multi-gig internet and fixing bottlenecks with the right Ethernet gear • planning a home network with VLANs, firewalling, access points, and a NAS media library • weighing whether to cancel streaming and build a personal Blu-ray rip collection • arguing Childish Gambino's catalog deserves a bigger spotlight • reacting to Summer Game Fest reveals and the rising cost of gaming hardware • calling out Bethesda for the lack of Elder Scrolls VI and meaningful new announcements • comparing publisher habits to Bungie and Destiny 2 live service fatigue • getting hyped for Nintendo Direct highlights and the return of arcade nostalgia • noticing World Cup teams struggling with U.S. heat • talking upcoming cons and swapping stories about Atlanta mishaps  Support the showhttps://www.carolinaotakus.com/

Industrial network protocols decide whether a machine talks or stays silent. Chuck from Horner Automation breaks down how they win, fade, and converge.Chuck has spent 36 years at Horner Automation and lived through what the industry once called the fieldbus wars. Before Horner became known for its all in one controllers, it spent a decade building specialty IO modules for GE Fanuc during the era of DeviceNet, SDS, InterBus S, PROFIBUS, and CANopen. His core argument is that most of those early protocols were technically fine. The ones that became standards won on the commercial weight of the companies backing them, not on superior specifications, with EtherCAT a rare exception that succeeded largely on technical merit.Trust is the recurring theme. Industry adopts slowly, and for years Ethernet was dismissed as too unreliable and not deterministic enough for control until Ethernet/IP, PROFINET, and Modbus TCP proved themselves. Today the market has settled around a big four set of protocols, and Chuck does not expect it to narrow further. For high speed motion he points to EtherCAT and PROFINET IRT as the implementations he most respects, since both step away from standard Ethernet at the device level to reach submillisecond timing.The episode is also a reality check on building your own hardware. Chuck and Dave describe how custom development routinely costs teams hundreds of thousands to millions of dollars, and how the real trap is obsolescence and maintenance rather than the first build. On the product side, the standout is FPD-Link, a serialization technology borrowed from automotive that carries video, touch, and power over one coaxial cable. Working with Safe Fleet, a maker of ambulances and fire trucks, Horner now mounts rugged displays up to seven meters from the PLC while still programming everything as one device.Looking ahead, Chuck argues that every PLC should now be treated as a data device first, because digitizing the process is the prerequisite for doing anything useful with AI. He also flags cybersecurity as the next burden for application engineers, with new mandates forcing both manufacturers and integrators to implement protections that were once optional. At Automate, Horner is showing HMI Connect and a 300 dollar CPU 151 that packs 18 IO points, wireless connectivity, and edge capability into a micro PLC.About Chuck and Horner AutomationChuck is a technical brand ambassador at Horner Automation, where he has spent 36 years across applications, product management, and education. An electrical engineer who started in the automotive industry, he now produces in depth tutorials on industrial protocols for the Horner APG YouTube channel. Horner Automation is a privately held controls manufacturer best known for its all in one PLC and HMI controllers, edge ready PLCs, and rugged hardware for industrial and mobile applications.Timestamps0:00 Introduction2:20 Chuck's Background and 36 Years at Horner Automation9:20 End User Engineer vs OEM Manufacturer Perspective13:20 New at Automate: HMI Connect and the CPU 151 Edge PLC21:30 The Fieldbus Wars and the History of Industrial Protocols24:20 What It Takes to Implement a Protocol Stack29:30 Why Protocols Win: Commercial Force vs Technical Merit32:40 Will Industrial Protocols Ever Converge?40:30 High Speed Motion: EtherCAT, PROFINET IRT, and Ethernet/IP44:40 FPD-Link: Rugged Remote HMI for Ambulances and Fire Trucks55:00 PLCs as Data Devices and the Push Toward AI1:02:40 Cybersecurity Mandates Coming for Application EngineersReferencesHorner Automation: https://www.hornerautomation.comAbout Your HostsVladimir Romanov is a co-host of The Manufacturing Hub Podcast and the founder of Joltek, an independent manufacturing and industrial automation consulting firm specializing in modernization strategy, digital transformation, and workforce development. Joltek works with manufacturers and investors to de-risk modernization and build the internal capability to sustain results.Connect with Vlad: https://www.linkedin.com/in/vladromanov/Want to go deeper? Vlad and the team at Joltek have covered related topics here:Understanding Plant Networks: https://www.joltek.com/blog/understanding-plant-networks-how-industrial-connectivity-evolvedIndustrial Ethernet Reliability: https://www.joltek.com/blog/industrial-ethernet-reliabilityDave Griffith is a co-host of The Manufacturing Hub Podcast and founder of Capelin Solutions, an industrial automation firm helping manufacturers adopt smart manufacturing technology. He brings 15 years of experience in industrial automation and digital transformation.Connect with Dave: https://www.linkedin.com/in/davegriffith23/Subscribe to Manufacturing Hub: https://www.manufacturinghub.liveLinkedIn: https://www.linkedin.com/company/manufacturing-hub-networkYouTube: https://www.youtube.com/@ManufacturingHub

Send us Fan MailPerfect-looking terminations and clean certification results can still hide the interference that slows networks weeks later. We break down EMI, RFI, and crosstalk in plain terms, then lay out the practical install rules that protect performance and your reputation. • why EMI is invisible but disruptive to Ethernet signals • common EMI sources on job sites, from lights to motors to transformers • the real cost of downtime and why customers only remember performance • how crosstalk differs from EMI and where it comes from • near-end, far-end, power sum, and alien crosstalk explained • termination mistakes that create marginal passes and failures • Cat 6A, higher frequencies, and the cable combing debate • prevention rules: separation, preserving twists, pathway fill, and bend limits • cable tray and bundling tips, including PoE heat concerns • shielding done right with proper bonding and grounding If you're watching this show on YouTube, would you mind hitting the subscribe button and the bell button to be notified when new content's being produced? If you're listening to us on one of the audio podcast platforms, would you mind leaving us a five-star rating? I'd love to hear from you in the comments below what's the worst EMI or crosstalk issue that you've ever encountered in a project. Drop your story in the comments below. Support the showKnowledge is power!  Make sure to stop by the webpage to buy me a cup of coffee or support the show at https://linktr.ee/letstalkcabling .  Also if you would like to be a guest on the show or have a topic for discussion send me an email at chuck@letstalkcabling.com Chuck Bowser RCDD TECH#CBRCDD #RCDD

Sun, 31 May 2026 15:00:00 GMT http://relay.fm/mpu/851 http://relay.fm/mpu/851 I Have Contraband 851 David Sparks and Stephen Robles David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. clean 4580 David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. This episode of Mac Power Users is sponsored by: Squarespace: Save 10% off your first purchase of a website or domain using code MPU. 1Password: Never forget a password again. Links and Show Notes: Credits The Mac Power Users Stephen Robles David Sparks The Editor Jim Metzendorf The Fixer Kerry Provanzano More Power Users: Ad-free episodes with regular bonus segments Submit Feedback Robot Assistant Field Guide Aqara Camera Hub G5 Pro Aqara Doorbell Camera G400 Robin Home HomePass for HomeKit & Matter HomeCam for HomeKit HomePaper for HomeKit Multi-State Sensor P100 – Aqara LLC Tailwind iQ3 Smart Garage Door Controller iSmartgate MINI THIRDREALITY Smart Garage Door Opener Govee Permanent Outdoor Lights Pro Govee Smart Cordless Table Lamp Classic IKEA launches new smart home range with Matter 5 Smart Home Upgrades I Should've Done Sooner - YouTube Aqara UWB Smart Lock U400 WITHINGS Body Smart Scale Immich Spokenly Introducing Shortcuts Playground - MacStories AI Built These Shortcuts - YouTube Stream Deck + XL | Elgato Prompter XL | Elgato MPU Timestamp Shortcut Audio Hijack Script Canisteo Motorized Blinds Roller Shade TRMNL | ePaper Dashboard DEVONthink 4.3 Herschel Menuwhere · Many Tricks Short Run — Sindre Sorhus DJI Osmo Pocket 4 Creator Combo DJI Mic 3 Bundle Shure MV7+ KU XIU Qi2.2 25W Magnetic Wireless Charger Anker Prime 3-in-1 Charging Station StealthTech Living Room Sound System | Lovesac Pixelmator Pro MacWhisper Fastmail MCP Server Superhuman DEVONthink TRMNL X Supercharge Elgato Stream Deck Elgato Key Light Air BetterTouchTool Keyboard Maestro Audio Hijack Bear Backblaze Parachute Carbon Copy Cloner Tailscale Snazzy Labs Hollyland Lark Wireless Mics Ulanzi RODECaster Pro 2

Sun, 31 May 2026 15:00:00 GMT http://relay.fm/mpu/851 http://relay.fm/mpu/851 David Sparks and Stephen Robles David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. clean 4580 David and Stephen answer listener feedback: rebuilding Apple Home with Aqara power-over-Ethernet cameras, smart scales, raw photo editing, connecting AI to email, off-site backups, the new TRMNL X display, and DEVONthink's MCP server. This episode of Mac Power Users is sponsored by: Squarespace: Save 10% off your first purchase of a website or domain using code MPU. 1Password: Never forget a password again. Links and Show Notes: Credits The Mac Power Users Stephen Robles David Sparks The Editor Jim Metzendorf The Fixer Kerry Provanzano More Power Users: Ad-free episodes with regular bonus segments Submit Feedback Robot Assistant Field Guide Aqara Camera Hub G5 Pro Aqara Doorbell Camera G400 Robin Home HomePass for HomeKit & Matter HomeCam for HomeKit HomePaper for HomeKit Multi-State Sensor P100 – Aqara LLC Tailwind iQ3 Smart Garage Door Controller iSmartgate MINI THIRDREALITY Smart Garage Door Opener Govee Permanent Outdoor Lights Pro Govee Smart Cordless Table Lamp Classic IKEA launches new smart home range with Matter 5 Smart Home Upgrades I Should've Done Sooner - YouTube Aqara UWB Smart Lock U400 WITHINGS Body Smart Scale Immich Spokenly Introducing Shortcuts Playground - MacStories AI Built These Shortcuts - YouTube Stream Deck + XL | Elgato Prompter XL | Elgato MPU Timestamp Shortcut Audio Hijack Script Canisteo Motorized Blinds Roller Shade TRMNL | ePaper Dashboard DEVONthink 4.3 Herschel Menuwhere · Many Tricks Short Run — Sindre Sorhus DJI Osmo Pocket 4 Creator Combo DJI Mic 3 Bundle Shure MV7+ KU XIU Qi2.2 25W Magnetic Wireless Charger Anker Prime 3-in-1 Charging Station StealthTech Living Room Sound System | Lovesac Pixelmator Pro MacWhisper Fastmail MCP Server Superhuman DEVONthink TRMNL X Supercharge Elgato Stream Deck Elgato Key Light Air BetterTouchTool Keyboard Maestro Audio Hijack Bear Backblaze Parachute Carbon Copy Cloner Tailscale Snazzy Labs Hollyland Lark Wireless Mics Ulanzi RODECaster Pro 2

PHP Podcast – May 28, 2026 Hosts: Eric Van Johnson & John Congdon Links from the show: PHP barely avoided disaster – YouTube CVE-2026-45793: Anatomy of a 14-Hour PHP Supply-Chain Near-Miss · graycoreio/github-actions-magento2 · Discussion #261 · GitHub An Update on Composer & Packagist Supply Chain Security PHP Tek: A Homecoming by Ben Ramsey Tek Roundup – Roave Speaking at PHP Tek 2026! #tech – YouTube PHP Tek is behind us, the ballroom is cleaned up, and we’re back to talk about all of it. Here’s what we covered: RIP Archie Bot After a long fight to keep him alive, Eric has officially retired Archie — the Discord bot built on OpenClaw that handled team standups, monitored PHP Architect’s Twitter/X group for join requests, and did a surprising amount of background work for the consulting team. When Anthropic shut down the OpenClaw API, Eric tried every model and service he could find to bring Archie back to form, but nothing got him all the way there. After a month of “almost working,” the call was made. He’s dead. Eric hasn’t ruled out revisiting it eventually — maybe with Claude Cowork — but for now, the bot is gone and the starting-soon link in Discord is broken because of it. Reviving a Six-Year-Old Codebase A client PHP Architect Consulting worked with from 2018 to 2021 has come back. The project — a reimagining of their app — was killed off when COVID hit and the CEO couldn’t align with the team’s vision. The last commit was six years ago. Now the client wants to bring it back, and Eric is spending the next few days analyzing what it’ll take to get it running again. Outdated packages, an old PHP version, and the general entropy of time are all on the checklist. Eric has genuine affection for this codebase — it was one of the first projects where he felt like the team was truly operating as a team, not just as an extension of him. Now it’s time to dust it off. Partner Spotlight: PHP Score → Our CVEs The PHP Score sponsor read may be getting a refresh — the folks at Artisan Build, who built PHP Score, have a new product they’re excited about: ourCVEs.com. It monitors your codebase’s Composer and NPM packages — and optionally your servers via a lightweight agent — for exposure to open CVEs, and alerts you when something needs attention. Pricing is generous: free forever for open source projects, $17/month for solo devs, $83/month for teams (or $1,000/year), with server monitoring scaling at $1 per server above 50. Ed from Artisan Build was at PHP Tek and made a strong impression. Go check it out at ourcves.com. How PHP Barely Avoided a Supply Chain Disaster Brent Roose released a 22-minute video covering a near-miss in the PHP ecosystem involving GitHub and Composer. The short version: GitHub changed their token format and briefly released it before Composer was ready to handle it. Composer was logging the token when the format check failed — meaning GitHub tokens were ending up in CI logs. In GitHub Actions, depending on how your action is configured, that container (and its token) might stick around for a while, giving an attacker a window to act. An alert developer caught the issue, used Claude to help research it, then did responsible disclosure — contacting the Composer maintainers and reaching out to Taylor Otwell, Vincent Pontier, and others in the ecosystem to disable their actions until the fix was in place. Update your Composer. GitHub rolled back the new token format but won’t keep it rolled back forever. Packagist MFA and Account Security Following up on the supply chain theme: Nils and Igor (Composer/Packagist maintainers) released a blog post on what they’re doing to improve supply chain security. The immediate ask for anyone publishing packages is to enable MFA on your Packagist account — it’s not required yet, but it will be. Eric went to check his own account, found MFA was already on, but noticed his username was still “diegodev” and he was using an old email. While updating it, he noted that Packagist didn’t require him to re-authenticate or confirm the change via the old email — a gap worth flagging if you have popular packages and someone ever gets into your session. PHP Tek 2026 Recap — The Good PHP Tek 2026 in Chicago is done, and despite everything (see below), the team is proud of how it went. Some highlights: Holly (CodeLorax) built a conference mobile app from scratch, released on both Google Play and the Apple App Store within 24 hours of the conference opening. The app let attendees build their own schedule, detected conflicting talk selections, sent push notifications when talks moved rooms, and even included a vendor lead-scanning feature where vendors could scan attendee QR codes to capture contacts. It was a genuine game-changer for the event. Eric and John named the conference elephant after Holly in appreciation — she also changed a trailer tire during setup, which sealed the deal. Clayton Kendall sponsored and produced the conference shirts and bags on an extremely tight timeline — shirts two weeks out, bags just one week before the event. Both were a hit. Attendees at the conference were getting questions about the rainbow PHP Architect shirt in particular. A job fair ran for the first time, with four companies represented. One hiring manager showed up even though they already had 1,400 applicants — because they knew that conference attendees are exactly the kind of motivated, self-improving developers they want. Attendees got to ask questions directly, including the real-world stuff like remote vs. office. Eric would love feedback on how to make it better next year. JS Tech debuted as a fourth track alongside the three PHP tracks, bringing in fresh faces from the JavaScript community. Eric came away energized by the cross-pollination — different people, different approaches to similar problems. Ben Ramsey and James Tickham (Rove) both wrote great blog posts about the conference. Ben’s will be featured in the magazine. Diana Pham also put together a video recap. Links in the show notes. PHP Tek 2026 Recap — The Incident On Monday during final setup, a hotel employee had a medical incident while walking through the main ballroom — leaving a trail that required hazmat-suited cleanup crews and forced the team to quarantine the ballroom, the hallway leading to it, and the adjacent bathroom. The person is okay and was back at the hotel by Friday, which was a relief. But in the moment, nobody knew what was happening or how long the room would be unavailable. The team had to rebuild the entire conference footprint overnight. The keynote moved, the JS Tech track went into the quiet room, vendors moved to the atrium, and the hotel staff — to their enormous credit — cleared their own furniture and accommodated every ask without complaint. Attendees were equally patient; once they understood the situation, there was no drama, just “tell us where to go.” The incident also took out the streaming setup for day one, compounding an already-difficult start. The solution that eventually worked — plugging the Ethernet into a hub before the streaming equipment — wasn’t tried until day three. Eric is mad at himself for thinking of it and not doing it sooner. PHP Tek 2027 — Save the Date (TBD) Planning for next year is already underway. The current target is April 2027 — away from the May timing that caused Eric to miss two of his kid’s band performances this year. Nothing is locked yet, but they’re working through venue and date options and hope to have an announcement soon. Links from the show: ourCVEs.com — Daily security audit on autopilot PHPScore — Technical debt monitoring for PHP Brent Roose — “How PHP Barely Avoided Disaster” (YouTube) Packagist — Enable MFA on your account PHP Architect Discord PHP Architect Merch Store PHP Architect YouTube Host: Eric Van Johnson X: @shocm Mastodon: @eric@phparch.social Bluesky: @ericvanjohnson.bsky.social PHPArch.me: @eric John Congdon X: @johncongdon Mastodon: @john@phparch.social Bluesky: @johncongdon.bsky.social PHPArch.me: @john Streams: Youtube Channel Twitch Connect & Hire PHP Architect Website Twitter/X Mastodon Hire PHP Developers Looking to hire PHP developers? Email support@phparch.com – Joe and the team are available for consulting, infrastructure work, Ansible playbooks, and code review. Partner This podcast is made a little better thanks to our partners Displace Infrastructure Management, Simplified Automate Kubernetes deployments across any cloud provider or bare metal with a single command. Deploy, manage, and scale your infrastructure with ease. https://displace.tech/ PHPScore Put Your Technical Debt on Autopay with PHPScore CodeRabbit Cut code review time & bugs in half instantly with CodeRabbit. Music Provided by Epidemic Sound https://www.epidemicsound.com/ Join Us Live Next Week Youtube Channel Got feedback? Join us on Discord at discord.phparch.com The post The PHP Podcast 2026.05.28 appeared first on PHP Architect.

“When you need these systems, they have to work 100% of the time,” says Jake Jacoby, CEO of TELCLOUD. “Our solution doesn't just meet the old copper standard — it exceeds it.” In part 35 of the TELCLOUD POTS and Shots Podcast Series, Doug Green, Publisher of Technology Reseller News, speaks with Jacoby about the hardware architecture powering modern POTS replacement and why reliability remains the most important requirement for life-safety communications. The discussion focuses on TELCLOUD's purpose-built POTScast 8 and POTScast 2 devices, which support eight and two analog lines respectively. Designed specifically for POTS replacement, the units support applications including fire alarms, elevators, emergency phones, security systems, fax lines, SCADA systems, and other legacy communications still dependent on analog connectivity. Jacoby explains that traditional copper phone lines historically delivered both dial tone and power directly from the carrier's central office, making them highly reliable during outages. TELCLOUD's approach replaces that infrastructure with a more resilient, modern design featuring battery backup, multiple WAN paths, LTE and 5G connectivity, and remote monitoring capabilities. Each POTScast unit includes a built-in 24-hour battery backup with optional expansion capability, along with support for multiple WAN connections including fiber, satellite, and cellular. TELCLOUD also supports Power over Ethernet deployments, allowing cellular routers from providers including Digi and ATEL to be placed up to 250 feet away from telecom closets where signal strength is stronger. Jacoby noted that TELCLOUD originally relied on existing analog telephone adapters but ultimately engineered its own hardware platform after determining that available solutions did not meet the company's performance standards for mission-critical deployments. “These devices are designed to sit in that telco room for the next 20 years,” Jacoby said. The episode also explores how TELCLOUD combines hardware, platform services, monitoring, field services, and channel support into a fully managed POTS replacement offering delivered through reseller partners globally. The “Shots” segment of the podcast featured Casa 1560 Private Selection Extra Añejo, a tequila aged more than three years in oak barrels and described by Jacoby as having notes of dark chocolate, dried fruit, and oak. For more information, visit telcloud.com or call 844-900-2270.

Simon Swords bootstrapped Fundipedia for nearly 20 years before selling to FE fundinfo in May 2025. He started in a garden shed with an Ethernet cable and a VoIP phone, ran three businesses at once for years, and eventually landed clients like HSBC, Barclays, and Legal & General. He negotiated the entire life-changing exit himself with his chairman and a ChatGPT subscription, without a corporate broker.In this episode, Simon and I go deep on what it actually costs to play the long game. We talk about the childhood that wired him to push through anything, why he says he would rather have died than given up, and why he cried the day the deal closed.We also get into why he sought therapy after the business was successful, not during the struggle. And why he thinks once you have the money, you're no longer allowed to be sad.This is a conversation about whether the exit actually sets you free, or whether the real work starts after. If you're playing the long game, this one's for you._______(02:00) Why bootstrapping for 20 years was the making of him(07:31) Cumulative childhood trauma and growing up in Dagenham(11:30) Anxiety as a superpower in business(12:59) "I would have rather died than given up"(15:38) Thinking he was having a heart attack on an onboarding call(18:30) His chairman buying him out for a million pounds(20:12) Waking up with the money and the anxiety still there(21:52) The dragon he was chasing for 30 years(23:30) Why most of his therapy came after success, not during(28:30) Running the exit himself with ChatGPT(32:31) What the cry on closing day was really about(34:56) Why once you have money, you're not allowed to be sad(38:30) The friends who anchor him to reality(42:23) The panic attack that made him stop(45:30) His 2026 plan to fill the space the business used to occupy(48:05) What he's most proud ofShow notes:Find show notes of each episode on ProfitLed.fm. Connect with our host:Follow Melissa on LinkedIn where she shares stories & lessons from her founder journey weekly.Connect with Melissa at melissakwan.com and subscribe to  'your founder next door', Melissa's weekly newsletter on what it's like to build a company without an abundance of resources and friends in high places.Follow @themelissakwan on Instagram and YouTube where she shares short videos of business advice and other truth-bomb sound bites.This podcast was brought to you by eWebinar:Find out how you can turn pre-recorded videos into interactive experiences with chat so you can run your demos, onboarding calls, and training sessions on autopilot, 24/7, without being there. Hop into a demo at eWebinar.com, no salesperson required.

Signal integrity engineers working on high-speed serial links, Ethernet, USB, PCI Express, and DDR memory interfaces need powerful simulation tools, but commercial software licenses can be cost-prohibitive. In this episode of the Altium OnTrack Podcast, host Zach Peterson sits down with David Banas, Solutions Engineer at Keysight, to explore three open-source Python packages that are transforming how engineers approach serial link simulation, IBIS-AMI model testing, and channel operating margin analysis. David walks through PyBERT, his most popular tool, demonstrating live how it handles eye diagrams, bathtub curves, jitter analysis, and equalization techniques like CTLE, DFE, and TX de-emphasis.   The pair explore PyIBIS, a Python tool for IBIS-AMI model developers, demonstrating its capabilities in analyzing signal transmission parameters. They look at how it helps debug models and apply equalization techniques to address signal distortion. Understanding this tool is crucial for effective data transmission analysis and ensuring signal integrity, especially when working with complex designs in Altium Designer or Cadence Design Systems.  

Wendy is back from hauling robots to Texas and getting ready to drive another one to California, so the crew leans hard into life on the road with Linux. Bill talks about moving his systems over to Bazzite, tells the story of an overworked NVIDIA 1080 that literally ate into another GPU, and explains how HomeBridge 2.0 keeps his smart‑home world humming. Nate shares his first impressions of Tux Manager, a Linux clone of the classic Windows Task Manager, and walks through the Framework‑plus‑Flip‑Go combo that makes his roaming setup feel like CubicleLabs away from home. From Steam Decks and One X Players to UniFi travel routers and noise‑canceling headphones, everyone opens their travel bags and talks about the gear they actually trust when Wi‑Fi is sketchy and power outlets are rare. Wendy also geeks out over her new MOVA V50 robot vacuum, complete with a dedicated “Sentinels” Wi‑Fi SSID, and how little self‑hosted comforts make a hotel room feel just a bit more like a homelab. Along the way, there are jokes about Ethernet‑cable hair, data having weight, and why the best layover is the one where your SSH tunnel actually connects. If you're curious about the recent Linux vulnerabilities and the ABCs of CVEs, don't miss SUDO Show 76, where they break it all down in a fun and informative way. Connect with the Hosts on Discord: Matt – @Dark1ltg Wendy – @Wendy.sh Nate – CubicleNate.com @CubicleNate Bill – @ctlinux on Mastodon Special Guest: Bill.

“These are purpose-built devices,” says Jake Jacoby, CEO of TELCLOUD. “They're UL listed, certified, tested, and designed specifically for this business.” In the latest episode of the TELCLOUD POTS and Shots Podcast Series, Doug Green, Publisher of Technology Reseller News, speaks with Jacoby about the hardware that makes modern POTS replacement possible. Jacoby showcases two TELCLOUD devices: the POTScast 8 LTE PC228 LTE, which supports eight analog lines, and the POTScast 2 LTE PC222 LTE, which supports two. Both are designed to support legacy and life-safety systems such as elevators, fire alarms, security systems, fax lines, SCADA applications, modems, and emergency phones as copper lines are phased out. The POTScast platform combines analog support with modern LTE and WAN connectivity, including broadband, Wi-Fi as WAN, satellite, and cellular. Each device includes 24-hour battery backup, helping ensure that critical communications continue even when building power fails. Jacoby also explains TELCLOUD's modular design. Because cellular signal is often weak inside telecom rooms, TELCLOUD supports Power over Ethernet, allowing routers from partners such as Ericsson, Peplink, Digi, InHand, ATEL, and Seego to be placed up to 250 feet away for better reception. The episode closes with the Shots segment, featuring Herencia Historico Grand Reserve Extra Añejo, a five-year-aged, small-batch tequila from Jalisco presented in a distinctive handcrafted bottle. For more information, visit telcloud.com or call 844-900-2270.

On this week's show a listener asks for some help with keeping his audio in sync with his video. We also discuss how to turn off the ACR on your Smart TV. But first we read your emails and take a look at the week's news. News: Here's What's Coming in the 2026 Apple TV Roku's Howdy Streaming Service Reaches an Estimated 1 Million Users Deal Alert! 65" TV for $238 Audio Sync in a Home Theater Byron's request for answers to some specific questions on audio sync: I'd appreciate it if you guys could provide some "guiding principles" on syncing audio in a home theater setup. I have four questions: 1. Should the AVR be the ONLY place to mess with syncing settings (when everything runs through it, including ARC)? Yes, in most cases—this is the recommended approach. Start with AVR settings at zero or Auto, enable Auto Lip Sync if available, and adjust the manual audio delay there. Avoid adjusting on the TV or sources unless you have a specific reason like a stubborn source that bypasses the AVR. Changing multiple devices creates conflicts and makes troubleshooting harder. 2. If AVR is the main adjustment point, do sources automatically stay in sync after setting it once?  Often yes, especially with Auto Lip Sync enabled and consistent sources. The AVR's delay setting (or per-input memory) applies across similar content. However: Different video formats, resolutions, SDR vs. HDR/Dolby Vision,  60Hz vs. 24p or processing modes can introduce varying delays. Some AVRs store audio delay per input, so one good setting per source/input often suffices.  3. For Fire TV Sticks, Apple TV, etc.: Do sync settings apply across all apps, or per-app? Fire TV Stick: The AV Sync Tuning (under Settings > Display & Sounds > Audio) is generally a device-wide offset. It should hold across apps for the HDMI output. Individual apps might have minor internal variations, but a global tweak usually covers most cases. Reboot the stick if sync drifts. Apple TV: No built-in manual per-app delay slider in standard settings. There's a Wireless Audio Sync calibration that uses the iPhone for measurement, which is more global. Different apps (e.g., Netflix vs. others) can sometimes show varying sync due to their decoding/processing—users often report needing AVR tweaks when switching apps. Match Frame Rate and consistent audio formats help stability.  In both cases, rely on the AVR for the heavy lifting. 4. Do higher-end AVRs allow different sync settings per input? Yes! Many mid-to-high-end models store audio delay/lip sync per input source. Examples include Denon models with "Master Audio Delay" or similar, where you can set and recall different ms offsets (often 0–500ms) for each HDMI input. This is a big convenience for multiple devices. Check your AVR manual for "Audio Delay," "Lip Sync," or "per input" settings. Additional Best Practices Minimize variables: Disable unnecessary video processing (motion smoothing, noise reduction) on the TV and AVR to reduce video latency. Use "Game" or "Pure Direct" modes where possible for lower lag. HDMI/ARC specifics: Ensure high-quality HDMI cables. eARC is better than ARC for bandwidth and sync negotiation. Power cycle everything (unplug) after big changes. Order of troubleshooting: AVR Auto Lip Sync → Manual AVR delay → Source device tweaks → TV audio delay (last). Test tools: Use built-in sync tests on your devices or YouTube "lip sync test" videos. The Most Effective ways to Circumvent Smart TV Spying Last week we talked about how your TV was spying on what you are watching. This week we discuss how to prevent that from happening. The following are the most effective ways to circumvent smart TV spying and related data collection, ranked from easiest/quickest to most thorough. These also help limit proxy network enrollment in shady apps. 1. Disable ACR Directly in TV Settings (Quickest First Step) Most brands let you turn off Automatic Content Recognition (and related ad/personalization features) without losing core picture quality. Do this on every TV: Samsung: Home button → Sidebar menu → Privacy Choices → Terms & Conditions / Privacy Policy → Uncheck Viewing Information Services (and Interest-Based Ads if present). LG: Settings → General → System → Additional Settings (or Advanced) → Turn Live Plus OFF → Also enable Limit Ad Tracking. Sony: Settings → Initial Setup → Disable Samba Interactive TV. Vizio: System → Reset & Admin → Turn Viewing Data OFF. Roku TV / Roku-based: Settings → Privacy → Smart TV Experience → Uncheck Use Info from TV Inputs. Hisense / TCL: Settings → System or Privacy → Disable Smart TV Experience or Use Info from TV Inputs. Amazon Fire TV: Preferences → Privacy Settings → Turn off data tracking options. After changing, restart the TV. Check the setting again after any software update, as it can reset. Also disable voice assistants, microphones, and cameras (cover them physically if needed). 2. Block Internet Access to the TV (Highly Effective) The simplest long-term fix: Prevent the TV from phoning home at all. Don't connect it to Wi-Fi or Ethernet in the first place. Or, on your router: Create a guest Wi-Fi just for the TV, then use firewall rules, parental controls, or MAC address blocking to stop all outbound internet traffic (while allowing local network access if you stream from a NAS/Plex/Jellyfin). Advanced: Use a tool like Pi-hole or AdGuard Home on your network to block known tracking domains. Pro tip: Many people report the TV works fine (or even faster) for HDMI inputs and local streaming when fully offline. External streaming devices handle all internet needs. 3. Use the TV as a "Dumb" Display Only Treat your smart TV like a big monitor: Connect all content via HDMI from a more private device (never use the TV's built-in apps). Recommended external boxes (in order of privacy-friendliness): Apple TV — Clean interface, minimal tracking, no aggressive ads. NVIDIA Shield or other local-media-focused devices.  Raspberry Pi or HTPC running Kodi/Plex for full local control. This bypasses the TV's OS almost entirely. 4. Go Fully "Dumb" (Most Private Long-Term Solution) Buy a true dumb TV or large computer monitor (no smart features, no Wi-Fi, no ACR). Options exist in smaller sizes or from brands like Westinghouse for basic panels. Pair it with an external streamer or your own computer/laptop via HDMI. Many privacy-focused users prefer this setup over any "smart" panel. Important reality check: Disabling ACR and blocking internet stops most viewing-data collection, but no method is 100% foolproof against every firmware trick or future update. The nuclear option—keeping the TV completely offline and HDMI-only—remains the gold standard for privacy.  

Whoops!  How did this get missed?!  Is 12vHPWR really that bad? Why aren't you mad enough about DDR and SSD pricing?!  Nvidia is BUYING ... well it's not HP.  All that "Don't track me bro" stuff is for naught, Sony sells you less and you'll buy it, and Linux stops being able to be loaded on just about anything.  You cannot wait for the segment on Ethernet cables, I know.  All that and more!0:00 Intro0:35 Patreon2:10 Food with Josh3:57 Yes, DDR5 and SSD prices are still insane7:30 NVIDIA is not buying Dell or HP9:40 Also, NVIDIA warranty payments up 1000 percent in 202512:23 NVIDIA N1 engineering board leak14:15 Desktop CPU sales tank for some strange reason17:19 Sorry, you are still being tracked21:24 Sony Bravia TVs losing some OTA functionality23:13 Copper Clad Aluminum27:57 Linux drops 486 support!30:20 MacBook Neo almost sold out34:12 (In)Security Corner44:50 Gaming Quick Hits48:50 Picks of the Week1:02:58 Outro ★ Support this podcast on Patreon ★

Take a Network Break! We commence with a red alert on FastMCP, and then debate whether Anthropic’s Project Glasswing is a marketing stunt or a reasonable response to the growing ability of AI models to find and exploit software vulnerabilities. Iran targets US OT networks, startup Aria Networks unveils Ethernet switches purpose-built for AI factories,... Read more »

Take a Network Break! We commence with a red alert on FastMCP, and then debate whether Anthropic’s Project Glasswing is a marketing stunt or a reasonable response to the growing ability of AI models to find and exploit software vulnerabilities. Iran targets US OT networks, startup Aria Networks unveils Ethernet switches purpose-built for AI factories,... Read more »

On this week's episode of Hands-On Tech, Mikah answers a listener's question about which connection method — Thunderbolt, 10GbE Ethernet, or a USB-C to 10GbE Ethernet dongle — offers the best performance and reliability for video editing with network-attached storage. Don't forget to send in your questions for Mikah to answer during the show! hot@twit.tv Host: Mikah Sargent Download or subscribe to Hands-On Tech at https://twit.tv/shows/hands-on-tech Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Club TWiT members can discuss this episode and leave feedback in the Club TWiT Discord. Sponsor: joindeleteme.com/twit promo code TWIT

Worum geht's? Im ersten Teil unserer Folge zu PC-Netzwerken haben Henner und Gunnar von der Entstehung des Ethernet berichtet und die kommerziellen und technischen Grundlagen aus der grauen Vorzeit hergeleitet. In diesem Teil geht es um die praktische Anwendung dieser Technologien für das vernetzte Spielen, vor allem das Spielen in einem lokalen Netzwerk, einem LAN. Produktions-Credits: Sprecher, Redaktion: Henner Thomsen, Gunnar Lott Audioproduktion: Sascha Blach, Christian Schmidt Titelgrafik: Johannes DuBois

Achtung: Dies ist Folge 1 von 2, Teil 2 erscheint in einer Woche. Worum geht's? In den frühen 1970er Jahren war ein Computer noch eine Rechenmaschine: hochspezialisiert, raumfüllend und keineswegs dazu gedacht, mit anderen Maschinen zu kommunizieren. Dass aus dieser Ausgangslage ein weltweites Netz entstehen konnte, verdankt sich einer Reihe von klugen Köpfen, glücklichen Zufällen und einer Technik namens Ethernet, die bis heute die Grundlage der meisten lokalen Netzwerke bildet – und damit auch der LAN-Party. Henner und Gunnar erzählen die Entstehungsgeschichte der Computernetzwerke: vom ALOHANet der Universität Hawaii, das Terminals auf mehreren Inseln per Funk verband, bis hin zu den Forschungslabors von Xerox PARC, wo Robert Metcalf Anfang der 1970er Jahre die Idee zu Ethernet entwickelte, inspiriert durch einen nächtlichen Zufallsfund im Bücherregal eines Kollegen. Sie sprechen über die verschiedenen Netzwerktopologien (Bus, Stern, Ring), über Koaxialkabel und den Vampire Tap, über den Namen Ethernet und seine Herkunft aus der griechischen Mythologie und vieles mehr. Der zweite Teil der Folge widmet sich dann dem, womit alles begann: der LAN-Party. Produktions-Credits: Sprecher, Redaktion: Henner Thomsen, Gunnar Lott Audioproduktion: Matthias Kuhlmann, Christian Schmidt Titelgrafik: Johannes DuBois

*The air you breathe, the light you see, and the water you drink have all been quietly sabotaged by modern life—here's how to fight back without ripping your house apart.* Episode Summary In this episode, we shift from last week's hormesis deep-dive to the second weapon in your ancestral mismatch arsenal: your environment. You'll discover how degraded air quality, flickering LED lights, contaminated water, and invisible EMFs are silently wrecking your sleep, fogging your brain, and loading you with toxins—then get a practical, budget-friendly environmental audit you can start this weekend. Question of the Day