The National Energy Research Scientific Computer Center (NERSC) in California just got a new supercomputer touted as the fastest in the world for Artificial Intelligence (AI) applications.
Named after Saul Perlmutter, an astrophysicist who discovered dark energy, the supercomputer will use the 6,000+ NVIDIA A100 Tensor Core GPUs on board to build the universe’s biggest 3D map to date.
According to NVIDIA, the Perlmutter supercomputer is currently capable of delivering four exaFLOPS performance on mixed-precision 16 and 32-bit math AI. With phase two, the processor’s performance will be enhanced with the addition of the second set of 3,072 dual EPYC 7763 CPU-only nodes from AMD.
Upon completion, the supercomputer is expected to rank among the top 5 fastest in the world, based on benchmark from High Performance Linpack (HPL). NVIDIA claimed Perlmutter could process 1 year worth of DESI data in a few days, instead of taking weeks or even months to complete.
The Perlmutter system will be put to use by more than 7,000 researchers, and dozens of applications were developed with the goal of improving mankind’s understanding of climate science and astrophysics.
It is hoped that once completed, the universe map will aid researchers in learning more about dark energy, the driving force of the universe’s accelerating expansion.