2 ай бұрын
2 ай бұрын
3 ай бұрын
3 ай бұрын
3 ай бұрын
3 ай бұрын
3 ай бұрын
3 ай бұрын
Пікірлер
@datacentermechtalk6852 2 күн бұрын
It’s literally impossible to be able to comfortably scale your data center for AI there’s no way. Unless you know the exact kilowatt draw on your servers and that’s impossible. How many servers will it take? Heisenberg‘s uncertainty principle applies to this conundrum of building AI data centers. How do you scale to something that has no scale? At this point you’re just building data centers with massive amounts of megawatt usage but have no idea what the end goal is for AI because it’s just in the beginning stages, so everything has to be variable just like the quibit. Whoever is reading this please correct me if I’m wrong. I’m currently building a hyper scale. AI data center. Everything is one of a kind. The only thing that I’m sure of is the max MW that the data center is rated for.
@user-qm5ho6fv6h 9 күн бұрын
Still got that '76 Accord ? 😊
@fibconetfttxsupplier2424 9 күн бұрын
AI technology has indeed driven the demand for data centers, which has become a reality in the optical communication industry. High-density MPO products for cabling are more suitable for the future development of data centers compared to traditional patch cords.
@GhostBear.Goldsmith 13 күн бұрын
ChatGPT suggested Ray for my project. I'm thoroughly impressed. Still learning it but wow. We're going to need bigger GPUs. It would be cool if we could do training across nodes instead of a single learner node with remote workers.
@frankl1 20 күн бұрын
Very interesting talk. Is any of the fine-tuned models, the source code and the incident/mitigation datasets available publicly ?
@arielshpitzer 23 күн бұрын
Ok gamut supposed to be rec 2020\2100... but what gamma ? HLG ? 2084 PQ ? couldn't find what's expected and what's working in an optimal way.
@otaviorl08 24 күн бұрын
Amazing presentation!
@vincentkila Ай бұрын
So how to improve the image brightness?
@alexryassky5775 Ай бұрын
What "AS" stands for?
@ornelio-chauque Ай бұрын
Thanks for sharing, Shuffle-sharding is definitely worth to test.
@dragonceo5911 Ай бұрын
China is so much advanced in self driving cars.
@nicoletamocanu2694 Ай бұрын
Thank you for sharing this great presentation!
@LibertypopUK Ай бұрын
the maths makes no sense on the network diagram, 16 GPU per rack for a single switch times by 192 = 3072 then times again by 18 zones gives you over 50K gpus?
@booleangray2225 2 ай бұрын
Shuffle-sharding is an interesting approach. Thanks for sharing.
@JeyParakh 2 ай бұрын
Really worth the keynote, very crisp and comprehensive.
@joeybluee 2 ай бұрын
Nice work!
@EnderWang 2 ай бұрын
Guess I have chosen a thesis topic with great potential, MOQT.
@pcbona 2 ай бұрын
16:11 Nice hidden joke...
@evair_marinho 2 ай бұрын
Excellent content! Congrats!
@perfectalgos9641 2 ай бұрын
At 7min:07second, I see MITA next generation Accelerator. Is it correct or it was a typo? It should be MTIA I think.
@pgee630 3 ай бұрын
All agreed but one more important point is to consider re-use of heat. We can't still stick to the legacy concept of rejecting heat to the ambient. With the high power AI load we have much more opportunities to capture the rejected heat for re-use.
@datacentermechtalk6852 2 күн бұрын
😂 You think hedge fund owned Hyper scale data centers are trying to capture heat loss? 😂They’re building 500MW data centers. The cooling load is 5.7 billion BTU’s using plate and frame heat exchangers for free cooling, during low ambient temperatures. The only money saved will be when they’re forced to generate their own power. Hopefully we’ll see mini-nuke campuses, that are exclusively for Data centers. At this point I’ve heard zero talk of PUE, only worry of how much the new servers will actually draw. Over the next 10 years you’ll see 10X Data centers, along with island mode laws. I’m not disagreeing with you, I too love to talk combined heat and power projects, but this is a race to information superiority. They don’t care about power usage efficiency.
@juliakanina 3 ай бұрын
Anton 🎉🎉🎉
@dolphinblue1698 3 ай бұрын
His company doesn’t know this guy is a misogynist..
@jamaicauptown 3 ай бұрын
Do you guys leverage kernel by pass technologies at the host like ofed ?
@foobars3816 3 ай бұрын
I'm quite surprised this isn't just the norm for international calls. It is what I always assumed would be the infrastructure setup. Does this mean until recently providers didn't make use of their backbone to improve traffic between users? The group call network just seems to be an extended star network, which is also similar to the hub and spoke model used by airlines. Users hop on a plane at an airport close to them, the plane flies to a central hub, the passengers jump onto another plane which takes them to the destination airport where they all them go off on their final leg alone. As mentioned, it is more strange that we are talking about this in 2024.
@foobars3816 3 ай бұрын
Shame it isn't actually released so we could verify these claims
@JordanPollard-cv6qd 3 ай бұрын
are all the calvin and hobbes comics subtle references to calvin transactions? 😆
@cosmosPK 3 ай бұрын
"Because of Meta's Backbone" :)
@cosmosPK 3 ай бұрын
Cheers to the next 10 years! There is so much work to be done!
@cosmosPK 3 ай бұрын
amazing times indeed!
@andreydung 3 ай бұрын
Amazing sharing!
@JohnnyDTruth 3 ай бұрын
Poor racist Pavel Punsky lost his job
@mandersen94 4 ай бұрын
Great talk Peter. Great seeing you share these important topics that are part of companies culture. Also like how you talked about unsubscribing problems. Sometimes we try to think we need to solve / manage a larger quantity of complex things when the solution is to focus on having to manage less numbers of complex things as you only have so much time to dedicate to them. Keep sharing your great insights. - Mark Andersen
@user-oq5ki8ml2r 4 ай бұрын
Great Speech. More people should have seen this.
@niggafagga 4 ай бұрын
Punsky the shlomo
@ronakshah725 4 ай бұрын
Amazing presentation!
@cosmosPK 4 ай бұрын
this is very good!
@EathanCano 4 ай бұрын
Thank you so much for your hard work! 😊 I have a quick question: 🤷♂️ I have a set of words 🤷♂️. (behave today finger ski upon boy assault summer exhaust beauty stereo over). Can someone explain what this is? 😅
@sidiouedraogo8324 4 ай бұрын
Great Talk !!!!
@uquantum 4 ай бұрын
Terrific summary and view of where we're probably heading with WebRTC in 2025 and beyond. Thanks Tsahi!
@sosexymagazinerobinwatkins8467 4 ай бұрын
I would like to speak with somebody personally about this workflow processes on Facebook I'm receiving messages about my business page saying they may shut me down because of logo and infringement but that logo is my logo which I've paid for. Can you help me with this situation? @Scale
@pingu-something 5 ай бұрын
talk is cheap...share examples. SD 1.5 lightning exist since long time
@seephor 5 ай бұрын
"it's very common for both old and new code paths to exist in the codebase simultaneously controlled by the use of conditional flags" Dear god what a fricken nightmare.
@zhangzhuo2636 5 ай бұрын
awesome！
@Caskeyboardproducer2023 5 ай бұрын
Amazing speech. Open source collaboration via the innovation of the internet and AI are incredible. Incredible keynote! Thanks for sharing!
@harryniedecken5321 5 ай бұрын
Thank you for the information. It would be really interesting if there was a way to use the rejected heat to purify water. An example is like naval vessels use vacuum distillation to purify sea water into fresh water.
@raymobula 5 ай бұрын
Lin’s videos popping up over the last weeks are so insightful. 😊
@saifal-badri 5 ай бұрын
Amazing work, thanks for the demo at 12:55. MLow sounds amazing much better than Opus 6kbps, although I was hoping for more comparisons with other low bandwidth codecs.
@CoolxZoid 5 ай бұрын
hello
@gokukakarot6323 5 ай бұрын
Why wasn’t an existing messaging queue like Kafka or rabbit used, with a redis backed heap task cache?