Eight projects to gain early access to the Frontier supercomputer

In preparation for the Frontier supercomputer , the US Department of Energy ’s (DOE’s) Oak Ridge Leadership Computing Facility (OLCF) has selected eight research projects to participate in its Center for Accelerated Application Readiness ( CAAR ) program.

Through CAAR, the OLCF will partner with application core developers, vendor partners, and OLCF staff members to optimize scientific applications for exascale performance, ensuring that Frontier will be able to perform large-scale science when it opens to users in 2022.

“CAAR is a collaboration,” said Judith Hill, group leader for the Scientific Computing Group at the National Center for Computational Sciences (NCCS). “The application developers tell us what’s broken or not available in the current software stack and what compiler and tool features are their priority, and that feedback helps us prioritize efforts with our vendor partners. These teams help set the direction of software development for this machine.”

As part of the CAAR program, teams will receive support from the US Department of Energy’s Oak Ridge National Laboratory Cray/AMD Center of Excellence staff, access to early generation hardware platforms, and early access to Frontier itself. Before Frontier’s delivery, each team will use Summit to further develop its projects and to establish a figure of merit, a metric determined by each team that will be used to measure some form of quantifiable performance increase. At the end of the CAAR program, the users will once again run their codes, this time on Frontier, with the results examined based on the individual figures of merit.

To be selected for the CAAR program, projects must show a high potential for scientific advancement that cannot be achieved on petascale computers like Summit, the OLCF’s IBM AC922 system, which is the fastest supercomputer in the world as of June 2019.

“One of the key components of every one of these projects is that they have a problem that only Frontier will enable them to solve,” said Bronson Messer, director of the CAAR program. “That’s one of the most important criteria for being in CAAR. They’re problems that, as recently as 5 years ago, would have seemed impossible. Figuring out what’s possible and then putting the flag out one step beyond that isn’t the easiest thing in the world, but I think we’ve assembled a team of people who can do that.”

Using Frontier, the eight teams will research a broad range of topics, from simulating mass outflows from galaxies like the Milky Way to understanding the way viruses like Zika enter host cells in the body. Some of the projects are set to create massive amounts of data and provide the perfect opportunity for achievements in deep learning and machine learning applications.

“With Summit there was a lot of excitement around machine learning and deep learning, but it was like this explosion of possible ideas,” Messer said. “Some of the Frontier CAAR projects have very specific, targeted needs where machine learning and deep learning can be brought to bear. The volume of these simulation results means you can’t just sit someone down to look at them over and over again, and that’s where something like a machine learning or deep learning approach could really help. I think we are starting to find the place where machine learning, deep learning, and artificial intelligence can really help scientific simulation.”

Frontier is slated to be more than 5 times faster than Summit, with an application performance upwards of 1.5EF, or 1.5 billion billion calculations per second. The system will make use of Cray’s new Shasta architecture and feature one AMD EPYC CPU and four AMD Radeon Instinct GPUs per node.

UT-Battelle LLC manages Oak Ridge National Laboratory for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science .