Building a GPU cluster for AI

Подписаться 4,4 тыс.

Просмотров 107 тыс.

50% 1

Whitepaper: lambdalabs.com/gpu-cluster/ec...
Learn, from start to finish, how to build a GPU cluster for deep learning. We'll cover the entire process, including cluster level design, rack level design, node level design, CPU and GPU selection, power distribution, storage, and networking.
This talk is based on the Lambda Echelon GPU Cluster whitepaper. The whitepaper can be found above.
Slides for the talk can be found here:
files.lambdalabs.com/How%20to%...
Errata:
- Slide 46 contains an erroneous diagram with a connection from the storage server to the compute fabric network, the storage server does not connect ot the compute fabric network. The correct diagram is available in the whitepaper.

Наука

Опубликовано:

2 авг 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 54

@onlooker365 7 дней назад

Ground level details with all the critical aspects covered nice for GPU Cluster to the last cable length calculation.

@fundyourhustle 25 дней назад

One of the best presentations on GPU cluster design, even at 3 years old. Great teaching skills!

@randahan215 11 месяцев назад

Extraordinary presentation. Covered all the important topics in depth and with real teaching talent. Many thanks!!

@randahan215 7 месяцев назад

Most professional and holistic explanation I heard about this topic. Thank you so much!!

@dr.mikeybee 10 месяцев назад

Thank you. You got me started years ago with your lambda stack -- the only way I could get TensorFlow installed on Linux.

@yassinebouchoucha Год назад

Thank you for highlighting an underrated topic/options that company should re-consider within their compute infrastructure.

@ilyboc 9 месяцев назад

Really good analysis and presentation!

@NSPK- 10 месяцев назад

Very expert suggestions for hpc and compute sizing.

@lovanda2000 25 дней назад

Best of the best presentation on server clusters. Author presented deep understanding of server clusters so that he explains things in an easy way. thank you!!!

@brianwesley28 3 года назад

Thanks for the video.

@carlschumacher5510 2 года назад

Its nice to see a holistic explanation of designing / building / installing a complex multi-rack system...As someone that has spent years working on both sides of the "analog/digital divide" (physical data center world / digital world's various segments), the un-sexy physical aspects of available rack space / power / cooling / floor loading / network uplink bandwidth are often overlooked (often assumed)...A semi arrives with a pallet: "Hey Carl, you can have this online in a couple days, right?"

@lambdacloud 2 года назад

Hey Carl, thanks for the kind comment. Glad you like the video. It's always funny how difficult it can be to 'bridge the divide' between the physical world and virtual world. Many SWEs expect to be able to "spin up" 1000 servers with an API call and forget that there are actual physical objects and tons of people that actually make that happen when you're on-prem.

@anatolystrashkevich7621 Год назад

very informative, thanks!

@peterxyz3541 Год назад

Thanks. I’m planning on building a “massive” 2 GPU system for home use.

@fundoo203 9 месяцев назад

How did it go man? I also want to build something like that and then stumbled on this video, which is excellent

@sanaullah-qureshi Год назад

very informative , thank you.

@uzairqarni7782 6 месяцев назад

This was amazing. Thank you.

@thePyiott 2 года назад

Great insight!

@user-us7oi6jw5i Месяц назад

Really complete, thank you!

@natexetan5732 3 года назад

thanks for the inspiration

@vtrandal 2 года назад

Excellent.

@austynr Год назад

Genius bait and switch. Props!

@metal_mo Год назад

Lambda needs an explanation on the difference between "building" and "designing".

@WorldMover 2 дня назад

What a remarkable video

@ProjectPhysX 3 года назад

Lots and lots of A100 GPUs. Every single one of them is a monster, almost 2x faster memory than the next best GPU. An entire room full of A100 racks... holy cow.

@programmingwiththotho4641 6 месяцев назад

Your are insane, thank you

@eyadmufti Год назад

it is a lecture more than a tutorial, Thx.

@cyberspider78910 2 месяца назад

Highly appreciated...RU-vid should have a separate category called Founder's video.

@Bloodycub666 2 года назад

I just love this kind things. How do i can start this kind bussnes how i can find customer for like small node and start building up

@glennisholcomb592 11 месяцев назад

I have three computers, and a nas, and a external hub. I think that I don’t need a another server because of the NAS. As far as my architecture goes, is there anything else that you can advise?

@petevenuti7355 11 месяцев назад

What if I have a model that I just want to run as provided, it hasn't really been optimized to run around the cluster and has memory requirements greater than any individual system I have. I feel safe to assume that for that specific case a shared distributed memory model would be the solution to run that specific app, yes? Is there any distribution of Linux that has support for such a memory model? It doesn't have to be a full-blown single system image. Perhaps a patch to the memory management driver so storage can be treated as an extension of system memory and not swap memory? Does any such software exist?

@HarishN.J 3 месяца назад

Hey Stephen, this is highly informative. I work on this clustering. Now am able to connect the dots and get the bigger picture. where can i read about the relationship between numa topology and GPU peering capability.

@ikbo 2 года назад

Do you guys have a gpu cluster optimized for 3d rendering.

@loadmastergod1961 2 месяца назад

I want to build a multi dual epyc 7742 based system for goofing around learning this stuff.

@chaoticblankness 7 месяцев назад

Very Based

@mengxu2026 3 года назад

Our group ordered around 10 lambda PCs 1 year ago. Right now more than 5 have problems. Some of them do not start up. Mine gets stuck randomly....

@yugr 3 года назад

Have you tried looking into the reasons?

@lambdacloud 3 года назад

Meng Xu, you can email support@lambdalabs.com 24/7 or call +1 (866) 711-2025 during business hours. Sorry to hear you're having issues, I'm sure we'll be able to resolve them quickly.