Hewlett Packard Enterprise Co and Advanced Micro Devices Inc on Wednesday said they had won a $600 million deal to deliver a supercomputer that will be used by the U.S. Department of Energy’s nuclear security arm to support the nation’s nuclear arsenal.
The computer, dubbed El Capitan after the famous rock face in California’s Sierra Nevada mountains, will be housed at the Lawrence Livermore National Laboratory. It will also be used by two other national nuclear labs, Sandia National Laboratories and Los Alamos National Laboratory.
The awarding of the chip contract to AMD is a big win for the company, which now has deals for two of the three most powerful computers under development by the U.S. government.
All but one of the world’s current top 10 fastest supercomputers use central processor chips from either Intel Corp or International Business Machines Inc, according to supercomputing research group TOP500. The exception is the third-fastest, which is located in China and uses a domestically developed chip.
Since the end of live detonation tests, using computers to simulate nuclear blasts has become a core part of weapons design. The enormous computing power required has also become a source of international competition and tension.
Of the world’s four fastest supercomputers, two are located in the United States and two are located in China, according to TOP500.
“That is why it is so important to push the edge” of computing technology, Bill Goldstein, director of the Lawrence Livermore National Laboratory, said at an event in San Jose, California announcing the news. “To make sure we’re there.”
The United States has barred U.S. firms from doing business with some Chinese supercomputer groups that U.S. officials say have Chinese military ties.
Pete Ungaro, senior vice president and general manager of high-performance computing and mission-critical solutions at HPE, said El Capitan will be about the size of two basketball courts and weigh as much as 35 school buses when complete.
The machine is likely to be 30% faster than previously expected thanks to the use of AMD’s forthcoming “Genoa” central processor units, or CPUs.
El Capitan will also use four AMD graphical processing units, or GPUs, for each CPU chip to speed up artificial intelligence tasks, they said. AMD makes both chips and a system for them to share computer memory, which is more difficult for systems that use chips from separate providers like Intel and Nvida Corp, analysts said.
“AMD is the first to market with them both in same package,” said Addison Snell, chief executive of Intersect360 Research.
HPE said it expects to deliver El Capitan in 2023. If it works as predicted, the machine will have a speed of 2 exaflops, meaning it will be able to perform 1 quintillion – or 1,000,000,000,000,000,000 – calculations per second and would be faster than the current top 200 supercomputers in the world combined. AMD and Cray last year won a deal for a 1.5 exaflop computer with the Energy Department, and Intel and Cray last year announced a supercomputer designed to run at 1 exaflop. (Reporting by Stephen Nellis in San Francisco; Editing by Edwina Gibbs & Shri Navaratnam)