DOE's Frontier serves as poster child of exascale -- for now

What are the killer apps envisioned by exascale supercomputers at the U.S. Department of Energy’s National Labs?  Try to pick just one.

“There are 24 first mover apps that span the DOE mission space, but frankly there will be hundreds to follow,” said Doug Kothe, lab director for DOE’s Exascale Computing Project and associate lab director for computing at Oak Ridge National Lab in Tennessee. He spoke during a virtual briefing on Thursday sponsored by the Semiconductor Industry Association.

Kothe ticked off an ambitious list of applications, large and small, that range from energy production to nuclear physics, astrophysics, risk assessments for earthquakes, precision medicine for oncology, assurance of the nation’s nuclear missile stockpile and more.

“It’s an incredible portfolio and these are not applications for one problem, but will deliver engineering to laptops and computers from Frontier, El Capitan and Aurora for years to come,” Kothe said.

Kothe’s office at Oak Ridge sits near Frontier, a leviathan that took top ranking in May as the world’s fastest supercomputer with 1.1 exaflops of performance, giving it the distinction of exascale, or quintillion calculations per second. Its theoretical peak performance is 2 exaflops. Frontier is an HPE Cray EX supercomputer that was built at a cost of about $600 million.  

Defending lofty applications and the spin-off effect of taxpayer-supported national labs and their supercomputers is constant effort, one that must go to Congress for regular appropriations, noted U.S. Rep. Chuck Fleischman, R-Tennessee. “Please remember the appropriations investment in supercomputing is important,” he said during the briefing. “We’ve made great progress from petaflops to exascale and the future is bright. I look forward to the accomplishments of this [Frontier] computer.”

Frontier and Oak Ridge National Lab are located in Fleishman’s district.  Over 12 years, he has voted to advance exascale investments in his role on the House Appropriations Committee.  When he started in Congress a dozen years ago, the U.S. was number one in supercomputing in the world, then slipped to number 5 and now is back in first place.  “This effort should be bipartisan and bicameral,” he added.

Ron Bewtra, director of leadership computing at Hewlett Packard Enterprise and a former NOAA official leading HPC work, said the U.S. needs supercomputing “to stay ahead of the expectations of Americans and our adversaries. We have to keep pace of American innovation. Supercomputers are more than a tool. “  Supercomputers were essential to advancements that led to the Webb Space Telescope, he added.

China has built two large supercomputers, Bewtra noted, but it is not clear how big they are or how they would measure up against Frontier and other U.S. supercomputers.

Bewtra and Mike Schulte, senior fellow for silicon engineering at AMD, said the public-private partnerships their companies and others have formed with DOE have led to innovations that both companies have used in chips and software used in their PCs and data center products.

Keeping energy in check used by supercomputers will be a constant challenge, but Frontier is doing its part, Kothe said.  Frontier was designed to be “used, useful and affordable and we’re there,” he said.  The original design called for using 20 megawatts per exaflop and Frontier is down to 15 megawatts per exaflop.  “We did it with the public-private partnership," he said.

Schulte said DOE and supercomputing innovations can perhaps help data center designers keep energy costs down. “The amount of power used by the data center is skyrocketing for cloud computing and AI,” he said. “It’s really important to make sure we are managing that power before it gets out of control.  For each megawatt used, it costs $1 million. Making them energy efficient is important for U.S. taxpayers.”

Kothe said DOE is in the midst of developing a revised roadmap for public funding of DOE supercomputing that should be ready by the end of 2022.  DOE is asking dozens of vendors for input in a formal request for information process.  “This is hard and a fast-changing field,” he said.

On the horizon for policymakers and appropriators: how supercomputers in the U.S. will be transformed from exascale to zetascale.  “It’s a hard to answer question,” said Jon Hoganson, corporate vice president for government relations at AMD and co-chair of the Task Force on American Innovation. “TFAI supports all research areas,” he said.

 RELATED: AMD fixes on $300B total addressable market with data center efforts leading the charge