News Intel Supercomputing 2023 Update: 1-Trillion Parameter AI Model Running on Aurora Supercomputer, Granite Rapids Benchmarks

[Admin]

Intel provided a slew of updates on its latest HPC and AI initiatives, including new information about its Granite Rapids Xeon CPUs, Max Series GPUs benchmarks against Nvidia's H100 GPUs, Guadi accelerators, and the company's work on the 'genAI' 1-trillion-parameter AI model that runs on the Aurora supercomputer.

Intel Supercomputing 2023 Update: 1-Trillion Parameter AI Model Running on Aurora Supercomputer, Granite Rapids Benchmarks : Read more

 

[weber462]

Support Petals AI. Support BOINC. Normal people should have access to AI. Normal people should have access to processing power.

 

[vanadiel007]

Support Petals AI. Support BOINC. Normal people should have access to AI. Normal people should have access to processing power.

No. It's only a matter of time before we see AI coin that can only be mined using AI neural networks.

 

[bit_user]

new Max Series GPU benchmarks against Nvidia's H100 GPUs

It's telling they only compared themselves to the H100 on non-AI workloads. For AI, they could only compete against the older A100.

 

[bit_user]

Support Petals AI. Support BOINC. Normal people should have access to AI. Normal people should have access to processing power.

As far as I can tell, Petals is only usable for inference. For training (and many HPC workloads), you need all that processing power to be very tightly-integrated with high-bandwidth interconnects and having fast access to huge amounts of data.

As far as normal people having access to it, there are cloud providers (including Nvidia) where you can rent time.

 

[jkflipflop98]

Support Petals AI. Support BOINC. Normal people should have access to AI. Normal people should have access to processing power.

Normal people don't have the electrical capacity to even boot up the amount of silicon it takes to run something like this. You basically need your own electrical sub-station to deliver the amount of energy this kind of thing takes.

Sure, in 20 years any old laptop at bestbuy will be able to do what these systems are doing, but for right now your best bet is to do as BU says and rent some cloud time.

 

[DougMcC]

I threw up a little bit when I read the page about the project written in Fortran.

 

[JayNor]

21,248 x 3
is number of PVC GPUs
2 CPUs and 6 GPUs per node.

 

[bit_user]

Normal people don't have the electrical capacity to even boot up the amount of silicon it takes to run something like this.

I think you missed @weber462 's point, which was to suggest that distributed computing could substitute for these kinds of supercomputers - not that an ordinary person would have even a single one of the Aurora-class machine in their home.

As I pointed out, distributed computing is only applicable to a rather limited set of problems. It's great when it does apply.

Sure, in 20 years any old laptop at bestbuy will be able to do what these systems are doing, but for right now your best bet is to do as BU says and rent some cloud time.

I'm not saying you're wrong, but I think there's probably not (yet) a technology roadmap that would get us there. More difficult than merely scaling compute performance is going to be improving energy efficiency to match.

Anyone interested in the subject should take a close read through this (especially the slides):

AMD Lays The Path To Zettascale Computing: Talks CPU & GPU Performance Plus Efficiency Trends, Next-Gen Chiplet Packaging & More

AMD talked about the future of computing, laying out its CPU & GPU trends in terms of efficiency & performance during ISSCC 2023.

wccftech.com

 

[bit_user]

I threw up a little bit when I read the page about the project written in Fortran.

I don't see where it says that in the article, but I would point out that Fortran has continued evolving, like C, C++, and other mature languages:

Fortran - Wikipedia

en.wikipedia.org

I've never used Fortran, but I'd be surprised if they hadn't added enough quality-of-life improvements to it, for it to be something I could live with.