Aurora with Intel delayed: US government orders new AMD Nvidia supercomputer

0
165

Intel's late delivery of all kinds of chips – Sapphire Rapids and Ponte Vecchio are simply not finished yet – prevents the Aurora supercomputer from finishing. Now an interim solution is to be found that relies on competitors with AMD Epyc and Nvidia's Ampere graphics cards, packaged in a system from HPE.

The U.S. The Department of Energy (DoE) is not just the Ministry of Energy, but also uses the computers to perform simulations for nuclear weapons tests. The supercomputer, baptized Polaris, should act as a quick interim solution and, according to Reuters, will be available this year, because AMD processors of the Rome and Milan types and A100 GPUs from Nvidia are available on the market. Here, however, the very high budget of 500 million US dollars and the client should also be helpful because the companies provide the appropriate hardware quickly and easily.

The Polaris with AMD Epyc will be connected to the network before the El Capitan ordered by the DoE for the year 2023 will be based on AMD Epyc “Genoa” with Zen 4 architecture and new Radeon Instinct graphics cards. El Capitan, which is also implemented by HPE, should achieve an output of two ExaFLOPS.

Update 08/25/2021 12:30 pm

Six minutes after the message appeared, the official press release wandered over the ticker. According to this, the information from Reuters is correct in advance, a mixture of AMD Epyc of the second and third generation is used, 560 in number, packed in 280 HPE systems.

It also contains a total of 2,240 Nvidia A100s, which shoulder the lion's share of the performance of up to 44 PetaFLOPS. The overall package should be four times as fast as the currently fastest solution of the supercomputers from Argonne National Laboratory (ANL).

The date is now only mentioned at the beginning of 2022, albeit the first parts should be delivered this month. The official announcement does not publish a price. $ 500 million, which was mentioned earlier, seems to be pretty high, especially since there is supposed to be an exascale system in the form of El Capitan for $ 600 million.