Energy Wants Software Sophisticated Enough to Run on Exascale Supercomputers

Marking a deliberate move to prepare for the approaching "exascale era" of next-generation supercomputers, the Energy Department announced plans to put forth up to $32 million to accelerate the development of advanced chemical sciences-centered software that can effectively run on the powerful, in-the-works machines.

The agency launched a funding opportunity this week for open-source codes that can be used to model and simulate complicated chemical phenomena—and can enable the complete use of its planned, future computing capacities. 

“The DOE national laboratories host some of the most powerful supercomputers in the world,” Director of Energy’s Office of Science Chris Fall said in a statement. “This research will harness these extraordinary capabilities to advance our understanding and control of chemical processes applicable to energy and a host of other areas.”

While Japan garnered the global top spot for the fastest supercomputer on the planet this year, as Fall noted America’s national labs are home to several next-level machines like Summit and Sierra, with immense computational power. Further, three U.S. exascale supercomputers—Aurora, Frontier and El Capitan—are presently being erected. 

On the general scale of system capabilities, exascale is next to be reached. Such sophisticated tools can perform a billion-billion operations per second and are a 10x order of magnitude more powerful than the fastest contemporary supercomputers. Energy officials noted in the funding opportunity, that they are now embarking on a new time when “chemical reactions and molecular systems can be controlled and matter can be built with atom-by-atom precision.” Computational models equipped to reliably forecast the behavior of molecules and materials based on theoretical calculation are at the heart of that kind of work.

But there’s an issue: Existing codes and software need to be adapted, or new ones need to be developed for the futuristic computing architectures. And Energy said such a move would be “essential for the U.S.” to compete in this realm.

“Open-source and commercial codes have established U.S. dominance in computational chemistry. However, that dominance is being challenged with the transition to complex heterogeneous architectures of high-performance computing platforms,” Energy’s document reads. “Many of today’s best chemical simulation codes are currently unable to efficiently use the capabilities made possible on existing leadership-class supercomputers.”

Through the new initiative, the department aims to support team-steered efforts to produce and refine open-source chemical simulation software that’s ready for advanced systems, and push forward the design of algorithms that can implement machine and deep learning, and other data-science approaches on exascale supercomputers. 

The opportunity is open to officials from national laboratories, universities, and industry, and Energy intends to make awards for both “large multidisciplinary, multi-institutional teams pioneering new software development, as well as smaller teams working to add functionality to existing software packages.” Up to $32 million could be dished out for various four-year projects, and the agency has $8 million in fiscal 2020 dollars and other funding contingent on congressional appropriations.

Pre-applications will be accepted until December 2 and responses to them will go out in January. The deadline for final applications is set for February 8.