The United States is now committed to building an exascale computer, some 30 times more powerful than today’s top machine. Yesterday, President Barack Obama signed an executive order creating a national strategic computing initiative, which aims to coordinate high-performance computing research and development between federal agencies. The order should make it easier for agencies to justify increasing their budget requests to Congress for supercomputing R&D.
“This is an extremely important step for high performance computing in the U.S.,” says Horst Simon, deputy director of the Lawrence Berkeley National Laboratory in California. The absence of a coordinated federal supercomputing effort had made it more difficult for agencies like the Department of Energy (DOE) to make their case with the Office of Management and Budget that they needed to boost supercomputing budgets, Simon says. The new order “reverses that.”
Jack Dongarra, a supercomputing expert at the University of Tennessee, Knoxville, says the new U.S. plans come partially as a response to concern over China’s surge in high-performance computing. According to the latest TOP500 list released earlier this month, the United States continues to lead the world in supercomputing power, with 233 of the top 500 machines. But that lead has been dwindling and is near a historic low. China has 37 of the top 500 supercomputers, but it has had the most powerful machine since June 2013. The world’s top supercomputer, China’s Tianhe-2, is nearly twice as powerful as the second place machine, Titan at Oak Ridge National Laboratory in Tennessee. And China recently announced plans to upgrade the Tianhe-2 machine to about 100 petaflops (quadrillions of floating point operations per second). An exaflop is 1000 petaflops.
Past U.S. supercomputing efforts have largely been pursued independently by different federal agencies, each looking to use advanced computing to help fulfill its mission, such as climate modeling and nuclear stockpile stewardship. But the top machines are now so complex and expensive that agencies must pool their R&D budgets. As a result, Simon says he and other U.S. supercomputing experts have been pushing the executive branch for years to come out with a plan for coordinating research across the federal government. The goal is break through obstacles that stand in the way of building much faster machines. (See this Science feature to learn more.)
Recently, several agencies have announced plans to integrate their supercomputing efforts. In November, DOE and the National Nuclear Security Agency—the part of DOE that oversees nuclear weapons—outlined joint plans to build an exascale machine by 2024. As a step along the way, DOE also announced last fall that it would spend $325 million to build two supercomputers between 100 and 300 petaflops
Reaching exascale is expected to be far more difficult than simply wiring additional computer processors together (as Science reported in this 2012 feature). Tianhe-2 already uses 18 megawatts of power, enough to power 18,000 homes. Using that same technology to get to exascale would require on the order of 540 megawatts, about the output of a nuclear power plant.
To get over such hurdles, Congress too appears poised to back an exascale push. Last week senators Lisa Murkowski (R–AK) and Maria Cantwell (D–WA) introduced comprehensive energy legislation that would direct the DOE to set up two parallel research efforts, each headed by a national laboratory in partnership with allied industry and academic groups. Because there are so many technical challenges to overcome in reaching exascale, the hope is that one of the two approaches would prove successful enough in the prototyping stage to warrant using it to build an actual exascale supercomputer. Computing experts largely back the twin track approach. The Senate energy committee today approved the legislation, but it still must be approved by the full Senate and House of Representatives. Thus far, however, Simon says that high performance computing enjoys broad bipartisan support in Congress.