In an effort to place the United States at the forefront of scientific computing, U.S. President Barack Obama has signed an executive order launching an initiative to build exaflop supercomputers.
The National Strategic Computing Initiative, or NSCI, will require supercomputers to achieve incredible new levels of performance and power efficiency. Two things GPUs are poised to deliver.
An exaflop supercomputer — roughly 30 times more powerful than today’s fastest machines — will need to rely on new technologies. One using CPUs alone would suck up 2 gigawatts of electricity. That’s the output of the Hoover Dam.
GPUs bend that curve. Unlike CPUs, they rely on large numbers of small, power-efficient computing cores that can handle up to 10 times more operations per unit of energy.
The payoff from machines efficient enough to reach exaflop speeds could be enormous. An exaflop computer would have the potential to provide unprecedented insight into the workings of human brain. Or lead to breakthroughs in personalized medicine.
GPUs Already Playing Key Role
The NSCI will extend efforts now underway at the U.S. Department of Energy, the Department of Defense and the National Science Foundation. GPUs play a key role. Last November, the U.S. Department of Energy unveiled plans to build two GPU-powered supercomputers.
Each so called pre-exascale system will deliver at least 100 petaflops of computing performance. One — the Summit system at Oak Ridge National Laboratory, designed for open science — is expected to be 150 petaflops. That’s more than three times the peak speed of today’s fastest supercomputer.
The United States isn’t alone in its bid to build faster supercomputers. China and other nations have exascale aspirations, but this move by the Obama administration is designed to help get the U.S. there first.
Building ever faster machines will require rethinking the way these machines are built. For example, one key new technology we’re developing, the high-speed NVLink interconnect, will help the CPUs and GPUs inside supercomputers exchange data 5 to 12 times faster. NVLink will play a key role in the next-generation Summit system at Oak Ridge and in the Sierra system at Lawrence Livermore National Labs.
A key focus for NCSI is developer productivity and portability. We recently announced a free toolkit around OpenACC, which dramatically simplifies programming for parallel processors, whether it’s x86 CPUs or GPUs. With OpenACC, the same code can run on different exascale architectures, delivering performance portability on any system.
The stakes are high. Scientists are already dreaming up new applications for a new generation of supercomputers. The European Commission’s Human Brain Project, for example, hopes to use exaflop computers to better understand the workings of the mind. And the White House’s own Precision medicine initiative is working to harness supercomputers to speed the advance of personalized medicine.
Exciting times ahead for supercomputing.