The US Department of Energy is close to an agreement to buy a supercomputer with Nvidia and AMD chips. Intel, meanwhile, is building an even bigger supercomputer for the office, but its delivery has been delayed for months, Reuters said, citing two people familiar with the case.
Nvidia and AMD’s Polaris an Intel-based Aurora machine designed for Argonne National Lab near Chicago, which would have been the fastest computer in the U.S. when it was announced in 2019. Instead, the Polaris, which will be up and running this year, will be a test machine for Argonne to begin preparing the software for the Intel machine, insiders said. The companies concerned did not wish to comment on the news
Intel, AMD and Nvidia are fighting for market share in chips used in data centers. U.S. supercomputers do scientific work for health, climate, and other researchers, and virtually test the country’s nuclear weapons. Key technologies that are pioneering systems will often seep into commercial data centers in the coming years, to the benefit of contracted chip companies.
At the first announcement of Aurora, Intel and Argonne said the machine It will be shipped in 2021, but Intel is still indebted to key Ponte Vecchio and Sapphire Rapids chips. Intel said in June that Sapphire Rapids chips would not be produced until 2022, but in recent days Will Moss, an Intel spokesman, said the company remained committed to shipping the computer in 2022.
Under the $ 500 million contract, Intel and its partners are required to supply a computer with so-called exaflop performance, ie 1 quintillion per second, ie 1 000 000 000 000 000 000 000 000 000 000. The first exascale computer in the United States is likely to be commissioned by another laboratory, Oak Ridge National Lab, Tennessee. This equipment will be built by Hewlett Packard with AMD chips and is expected to ship later this year.
According to sources, the Polaris machine used by Argonne will not be as powerful as Intel’s computer. Based on Nvidia’s A100 chips and AMD’s Rome and Milan chips, Polaris will be able to perform some exaflop calculations, but will mostly run slower.
Hardware, software, tests, curiosities and colorful news from the IT world by clicking here!
