The fastest supercomputer in the US today is Titan (currently third fastest in the world). Located at the Oak Ridge National Laboratory in Tennessee, it utilizes a hybrid architecture consisting of AMD CPUs and Nvidia GPUs to offer 20+ petaflops of performance requiring 8.2 megawatts of power. That may be fast today, but the future is exascale supercomputers, which achieve 1,000+ petaflops of performance.
The Department of Energy (DoE) realizes exascale supercomputers are "critical for U.S. leadership in areas such as national security, manufacturing, industrial competitiveness, and energy and earth sciences." But energy efficiency is of great concern. If Titan requires 8.2 megawatts to achieve 20 petaflops, imagine what 1,000 petaflops would require without some major breakthroughs. So the DoE created the PathForward program to focus on energy efficient exascale computing research.
Last week, the DoE chose six technology companies to receive $258 million of funding over a three-year period. Those companies are AMD, Cray, HPE, IBM, Intel, and Nvidia. Each company will also add at least 40 percent additional funding, taking the three-year total investment to $430 million split between hardware, software, and application development research.
The overall goal is to see a huge increase in computing power over today's best supercomputers (50x increase) without a huge increase in energy consumption. Nvidia gets more specific, stating the DoE's ambitious goal is, "to achieve exascale performance using only 20-30 megawatts." The company also points out that attempting to achieve an exascale computer with CPUs alone would take gigawatts of energy.
This isn't a new initiative for the DoE. AMD was awarded a $32 million grant back in 2014 to research exascale computing. There's also competition from China to consider. The two fastest supercomputers in the world reside in China (Tianhe-2 and Sunway TaihuLight). China has also promised to have a prototype of an exascale supercomputer ready before the end of 2017.