SAN JOSE, Calif. — The U.S. Department of Energy awarded $258 million to six tech giants, paving the way to build at least one exascale-class supercomputer by 2021. AMD, Cray, Hewlett Packard Enterprise, IBM, Intel, and Nvidia received three-year grants and will, in turn, invest $172 million in research for the PathForward program.
The grants are the latest phase of the project launched last year that has, so far, spent about $100 million, mainly on software development. The awards come at a time when China holds the top two positions on the list of the world’s Top 500 supercomputers and claims that it will deliver an exascale-class system in 2020, a year ahead of the U.S.
“This is quite an important milestone for the national exascale project … I would not characterize this as an effort to catch up with China,” said Paul Messina, director of the Exascale Computing Project and a program manager at Argonne National Lab.
The U.S. project aims to deliver a system that can provide sustained exascale performance on actual DoE workloads. The Top 500 supercomputers typically are measured by peak performance on the Linpack benchmark.
The U.S. systems aim to be 50 times more powerful than the biggest supercomputer in the country today, Titan, a 17-petaflop system at Oak Ridge National Lab. Titan is the world’s third most powerful system, behind 93- and 33-Pflop systems in Wuxi and Guangzhou, China.
At least one system should be built in 2021 according to the group’s timeline. Click to enlarge. (Image: Exascale Computing Project)
One researcher working on the project said that the original goal was to deliver an exascale system in 2023 that would consume about 20 MW. Bringing the date in two years may result in loosening the power limit to 30 MW, he said.
Meanwhile, China is building three prototypes for exascale systems that it will demonstrate later this year, said Jack Dongarra, a professor of engineering at the University of Tennessee and co-author of the Top 500 list.
One system is a follow-on of the Sunway Taihu Light that uses China-designed and -made processors. One from China’s National University for Defense Technology is reported to use ARM processors and China-designed accelerators, and a third system is being built by Sugon using a version of x86 processors licensed last year in a joint venture with AMD.
The DoE has estimated that it is investing as much as $5 billion in advanced supercomputers and their applications. However, some suggest that other branches of government could be involved in a larger, more coordinated effort.
“I think [that] the U.S. could be investing more deeply, but the investments they are making are making a difference that’s substantial and effective,” said one researcher.
For example, the National Science Foundation’s supercomputing centers could be upgraded to run at performance levels similar to DoE’s systems. However, President Trump has recommended in his budget that the NSF take a 10% budget cut.
The six companies will use the PathForward grants to accelerate R&D on a wide range of component technologies needed to build an exascale-class supercomputer. Their work spans microprocessors, memories, interconnects, packaging, and software.
The grants will not pay for the systems themselves. “Funding for acquisition and citing of the systems, as is usually the case, will come from the budgets of the national labs,” said Messina. “How much they cost hopefully will be reduced a bit by this [research] funding, but the exact number is a moving target.”
Next page: AI plays prominent role in research