The Tianhe-2 supercomputer developed by China’s National University of Defense technology is the most powerful system in the world with a peak performance of around 54.9 petaflops — the level below exascale — according to TOP500, a project than tracks the performance of supercomputers.
“We’ve allowed much of our technology to go overseas and be exported — so now we have to buy back technology we’ve developed,” said Sterling. “If we’re not careful, we will lose further ground.”
Yet the United States is still in the race. The Titan supercomputer at the Department of Energy’s Oak Ridge National laboratory is currently world’s second most powerful. And earlier this year, the agency struck a deal with Intel and supercomputing company Cray to deliver a system capable of 180 petaflops by 2018.
But supercomputing comes with its own set of challenges, including how efficiently they operate. Some existing supercomputer systems essentially waste the vast majority of their processing power when attempting to complete tasks, said Sterling, and that wasted processing power has real world energy costs.
“If you scale current technology up to exascale levels, it would be up to the range of a nuclear powerplant just to run one computer,” Binkley said. Addressing that problem will be one of the focuses of the initiative, he said.
And even if a machine with this sort of processing power is created, there remains the challenge of getting it to do what researchers want, said Sterling. “The overwhelming challenge is how do you program these machines,” he said.