Intel Corp. has been selected as the prime contractor to supply
what could be the world's fastest supercomputer, collaborating with
Cray Inc. on a $200 million Energy Department project for Argonne
National Laboratory that uses some previously undisclosed chip
technology.
The planned system, which won't be delivered until 2018, is the
latest sign that government agencies are investing to bolster U.S.
leadership in scientific computing amid stiff competition from
China and other countries. The machine at Argonne, a facility near
Chicago, marks the third phase of an Energy Department program that
kicked off in November with plans to spend $325 million on
supercomputers at two other national laboratories.
Supercomputers, referring to room-sized computers used to solve
scientific problems, have long been used at national labs for
computation-intensive chores such as simulating nuclear explosions.
A system in China in 2013 won the top spot on a twice-yearly
ranking of the world's most powerful computers, prompting calls by
U.S. scientists to boost funding to develop faster machines.
Franklin Orr, the Energy Department's undersecretary for science
and energy, called the Argonne machine, dubbed Aurora, an important
step for science. "It allows us to really work on the very tough
advanced simulation processes that underlie all kinds of energy
technologies," he said in advance of a press briefing Thursday.
Aurora is expected be the first supercomputer based on a design
called Shasta being developed by Cray, a longtime supercomputer
supplier to government agencies and companies. Intel will supply
processing chips and other technology.
Aurora will be available for use by researchers from companies
and universities as well as from national laboratories, Mr. Orr
said. Likely applications include developing more efficient solar
panels and wind turbines, quieter engines and better biofuels, the
Energy Department said.
The new system is expected to perform 180 trillion scientific
calculations per second, which would make it five to seven times
more powerful than comparable systems now in use. Though projects
designed to leapfrog Aurora are possible, the system should be a
contender to set speed records in 2018, said Barry Bolding, a Cray
executive.
Intel chips have been a mainstay of supercomputers since the
1990s. But the company hasn't served as a prime contractor in
delivering a supercomputer for nearly two decades.
Besides Intel's Xeon chips, which are used in conventional
servers, the company has been selling a specialized variant called
Xeon Phi that has a larger number of calculating engines. That
technology was originally designed to work in tandem with
conventional Xeons, but some new supercomputers use Xeon Phi chips
alone.
The Argonne announcement marked Intel's first public discussion
of a suite of technologies it calls the HPC scalable system
framework. Besides Xeon Phi, which will power Aurora, other
components include a new communications technology for connecting
chips, a new memory chip design and new file storage software,
Intel said.
Dave Patterson, president of Intel's federal unit, said the
framework was designed to deliver a balance of high performance and
low energy consumption while scaling from individual users to the
largest systems. Though Intel opted to take a prime contractor role
in this case, he said the company more commonly would sell
components to computer makers.
"This is not a change in the business model," he said.
Intel and Cray were also selected by the Energy Department to
supply a much smaller computer to Argonne in 2016. That system is
designed to help researchers start converting their software to
work on the Intel chip technology.
Besides systems based on Intel chips, some national labs have
opted for designs that use chips from Nvidia Corp., Advanced Micro
Devices Inc. or International Business Machines Corp.
Write to Don Clark at don.clark@wsj.com
Subscribe to WSJ: http://online.wsj.com?mod=djnwires