Nvidia is partnering with chip designer ARM. At the International Supercomputing Conference, Nvidia presented this partnership as a new path for the industry to achieve highly energy-efficient, AI-enabled exascale supercomputers.
Nvidia says it will make its entire lineup of AI and HPC software available to the ARM ecosystem by the end of this year. The company reports that after completing stack optimization, it will support all major CPU architectures, including x86, Power, and Arm. According to the CEO of Nvidia, Jensen Huang, supporting the ARM’s CPU architecture is very important, given the open architecture and its energy efficiency.
Huang describes that achieving exascale supercomputing “will significantly push the frontier of human knowledge.” He argues that supercomputers are essential for scientific discovery, but that power requirements will limit all supercomputers once traditional compute scaling ends. According to Huang, the combination of Nvidia’s Cuda and ARM’s energy-efficient architecture will give the entire hpc community a big boost.
In the most recent edition of the Top500 list, 156 is a supercomputer based on the ARM architecture. It concerns Astra, an HPE supercomputer used by the American Sandia National Laboratories; this one has been on the list since November 2018. In the HPL benchmark used for the Top500 list, the system achieved a performance of 1758 teraflops and a peak of 2209 teraflops.