December 2021

New Frontier

The Frontier supercomputer at the Department of Energy’s Oak Ridge Leadership Computing Facility (OLCF) will greatly exceed its predecessors in capability and speed. Hewlett-Packard Enterprises and semiconductor company AMD designed Frontier with DOE and Oak Ridge National Laboratory, OLCF’s home. The national user facility received the first hardware for Frontier in late July 2021 and had planned to make Frontier available in early 2022, says Justin Whitt, OLCF’s program director. “At this point we’re 100 percent on track to do that.”

Frontier is the latest in a line of record-breaking OLCF machines. Jaguar surpassed the petascale barrier in 2009, performing more than 1 quadrillion calculations per second. Titan outdid Jaguar in its 2012 debut and ran up to 27 quadrillion calculations per second. By 2019, the Summit supercomputer had eclipsed them all at speeds up to 200 quadrillion calculations per second.  Soon the Frontier exascale supercomputer will achieve more than 1.5 quintillion calculations per second.

 Making Frontier a reality has involved deep partnerships across multiple DOE laboratories, academic institutions and industry partners, Whitt says. “You’re talking about thousands of really smart people working to make this happen for years ahead of time.”

 The Frontier design process kicked off in 2018. DOE gathered about 150 experts from across its complex who assessed technologies and evaluated which proposals should be considered for building the supercomputer.

In addition, Frontier has been designed carefully to make its capabilities and software look like Summit to ease users’ transition to the system. “A lot of the programming and software environmental tools that will be available on Frontier are already available on Summit.”

Frontier will boast some novel technologies that are under nondisclosure agreements with their manufacturers, Whitt says. “The AMD family of processors, both the CPU and GPU that will be in Frontier, are very high-end, customized products that are meant for traditional high performance computing workflows as well as artificial intelligence programming and learning workflows.” And like Summit, the system should excel at simulation, data analysis and AI workflows.

“We’re honored to carry the exascale flag for the nation at this point,” Whitt says, adding it will be “uniquely rewarding when we start to see the kinds of science and engineering that are produced by these systems.”