News AMD Wins El Capitan: EPYC Genoa and Radeon Instinct to Power Two-Exaflop DOE Supercomputer

[Admin]

AMD wins the El Capitan supercomputer contract with its Zen 4 EPYC Genoa CPUs and Radeon Instinct GPUs.

AMD Wins El Capitan: EPYC Genoa and Radeon Instinct to Power Two-Exaflop DOE Supercomputer : Read more

 

[velocityg4]

I can't get past the name. Each time I read it. I just think this will be the most EPYC hackintosh ever created.

 

[derekullo]

Every time I read it I hear

O Captain! My Captain!

 

[BT]

Nvidia DOES NOT Win DOD/DOE contracts because they make their products in Taiwan Fabs and not in the US. IBM, Intel and AMD make their stuff in US Fabs.

 

[Deicidium369]

Uh yeah, no. AMD makes it's processors at the same place Nvidia makes their GPUs - Taiwan - that's the T in TSMC. The older Ryzen are made in GloFo's US facility - but the "7nm" is in Taiwan. Taiwan may not consider itself a part of mainland China, but mainland China disagrees.

 

[PaulAlcorn]

Intel is also reliant upon test and package facilities in China, among other pursuits. They have opened a new facility in Vietnam and Costa Rica to sidestep some of that, probably due to the trade war. However, most vendors are exposed to the Chinese supply chains in some form, AMD included. Doesn't matter where you make the chips if some of the key ingredients are sourced from China.

Note that Nvidia has won plenty of DOE contracts in the past.

 

[mihen]

I think Instinct MI60 verse GV100 shows two similarly specced cards, but the AMD product meshes better with the other hardware, it's newer, and it was probably significantly cheaper.

 

[Deicidium369]

Most of the activities in China are for foreign markets.

"Nvidia DOES NOT Win DOD/DOE contracts because they make their products in Taiwan Fabs and not in the US. IBM, Intel, and AMD make their stuff in US Fabs. "

is the comment I was responding to - the argument that Nvidia doesn't win because they fab in Taiwan is false - was just pointing out that AMD also makes their CPUs and GPUs at the same company in Taiwan.

Of course any large multinational like Intel has processes spread across the globe - I do, however, however, think that most of the work done on products does not come to the US market. Not sure what you mean by key ingredients - not the sand, not the wafers, not the fabrication, and maybe only raw materials to make the packaging are from China - the fabrication is done by machines from the EU and US. But, again, hard for any industry to not have some linkage to Chinese suppliers.

Supercomputing would not be where it is today without Nvidia GPUs - prior to that, it was massive amounts of CPUs. Intel's Larrabee derived products came at the exact wrong time - just before Nvidia GPUs started taking all of the oxygen out of the market - it is also the reason for Xe HPC - no need to cede a significant portion of the BoM to another company.

This whole manufactured trade war is total <Mod Edit>, amazing we have a president who thinks that raising tariffs on imports means there is money flowing into the treasury. Giving China MFN status was a MASSIVE mistake - but not understanding how tariffs work is the biggest blunder.

 

[Deicidium369]

I think Instinct MI60 verse GV100 shows two similarly specced cards, but the AMD product meshes better with the other hardware, it's newer, and it was probably significantly cheaper.

Meshes better? in what way? Newer is meaningless, cheaper is meaningless - what matters is ecosystem - Nvida's CUDA ecosystem is robust and comprehensive - and AMD's not even close - When XE HP/HPC drops one of the biggest challenges will be for Intel to build out tools to make migration from CUDA streamlined and compelling. Intel has the market muscle (installed base and $) to challenge Nvidia - at this point they are the only company that stands a chance to take on Nvidia.

 

[mihen]

I am talking about the communication between the Epyc processors and the GPUs. For someone like the DOE an existing ecosystem is not that important since everything will be made specific for the DOE.

 

[ta152h]

This is just a bad article.

Nowhere is Hewlett-Packard Enterprise mentioned, which by the way, owns Cray. Odd. I guess the author didn't know.

Also NVIDIA does have a very tightly knit GPU/CPU architecture, and had it long before Intel or AMD. Look up OpenPOWER to learn more. IBM/NVIDIA broke that ground years ago. First with OpenCAPI, then with NVLINK. Why the HP was chosen, I'm not sure, but it was not because NVIDIA did not have an intimate relationship with CPUs.

 

[Chung Leong]

Nvidia DOES NOT Win DOD/DOE contracts because they make their products in Taiwan Fabs and not in the US. IBM, Intel and AMD make their stuff in US Fabs.

Nah. Nvidia probably wasn't picked because after the acquisition of Mellanox it's now a direct competitor with Cray.

 

[levijonesm]

This is just a bad article.

Nowhere is Hewlett-Packard Enterprise mentioned, which by the way, owns Cray. Odd. I guess the author didn't know.

Also NVIDIA does have a very tightly knit GPU/CPU architecture, and had it long before Intel or AMD. Look up OpenPOWER to learn more. IBM/NVIDIA broke that ground years ago. First with OpenCAPI, then with NVLINK. Why the HP was chosen, I'm not sure, but it was not because NVIDIA did not have an intimate relationship with CPUs.

Do you also complain when Alphabet is never mentioned in articles about Google?