Yesterday’s New York Times reported that Chinese high-performance computing researchers have achieved a performance of 2.5 petaflops using their Tianhe-1A supercomputer. The Tianhe-1A machine is now the world’s most powerful computer, 40% faster than the fastest American machine located at Oak Ridge National Laboratory.
In an op-ed I wrote for the San Francisco Chronicle in June of this year, I argued the importance of maintaining U.S. leadership in high-performance computing. Of the top 500 supercomputers in the world, more than half are in the U.S., and 90% were built by U.S. hardware vendors. In the context of yesterday’s news, I’d like to give you a quick update on the Energy Department’s activities to strengthen U.S. leadership in this area.
I am spearheading planning within the Department to develop the next generation of supercomputers over the next decade, which will be capable of exaflop-class performance (a factor of 1000 more powerful than today’s most powerful computers). These machines will require paradigm shifts in both hardware and software in order to use significantly less energy per unit of computation and be more resilient against hardware failures. High-performance computing is a technology in which each generation of performance relies on the last, and we will use our most powerful computers to design and simulate the next generation. Hardware innovation for exascale machines will also be directly applicable to commodity electronics, making portable computers and smart phones much more powerful.
My office recently held a Simulations Summit, where I hosted more than 70 leaders of academia, industry, government, and national research laboratories to discuss policies and plans for bringing science and simulations to bear in support of national competitiveness. During his keynote address at the Summit, Secretary Chu said that “the DOE strategy should be to make simulation part of everyone’s toolbox.” We have established a website for the Summit , where you can find a summary of the event as well the presentations from several of our speakers. Industrial computing users at the Summit were enthusiastic about DOE’s plans to pursue exascale capability, as they understand that the hardware and software innovations necessary for this leap in power will also allow them to increase the pace of their innovation.
While the Chinese Tianhe-1A machine has been built using U.S.-designed processors and memory, it uses Chinese-designed interconnects that are faster those used in U.S. supercomputers. The Chinese have also become skilled in processor design and fabrication and are building another petaflop-class supercomputer using entirely indigenous components that is expected to be complete within the next 12 to 18 months. The computing capability represented by these new machines will provide Chinese consumers of high-performance computing with world-class facilities. It is clear that DOE will not be the only organization working to push the limits of computer performance.
Leadership in high-performance computing is a matter of both hardware and software. DOE leads the world in our ability to use high-end computers for both scientific and defense applications. We are also pushing hard to apply high-performance computing to our energy mission, and we have recently launched energy system simulation programs through our Energy Innovation Hubs for nuclear energy and energy-efficient buildings. As our Simulations Summit reinforced, driving the leading edge of both hardware and software also benefits U.S. industrial competitiveness. At the conclusion of our Summit, it was evident that a golden moment has presented itself to continue U.S. leadership in simulations, but concerted action and continued DOE leadership are necessary to turn this opportunity into reality. Yesterday’s announcement reinforces the need to act decisively and promptly to develop the next generation of U.S. supercomputers while harnessing existing computing power to drive innovation in science, industry, and national security.