World's fastest AI supercomputer is powered by 6000 NVIDIA A100 GPUs

NVIDIA has announced that the world's fastest AI supercomputer has been turned on. The machine is called Perlmutter, and it was officially dedicated on May 27 at the National Energy Research Scientific Computing Center. The supercomputer will deliver nearly four exaflops of performance to more than 7000 researchers.

Four exaflops of AI performance makes Perlmutter the fastest system on the planet on the 16-and 32-bit mixed-precision math AI uses. NVIDIA notes its current system performance doesn't include a second phase coming later this year to the system, which is based at the Lawrence Berkeley National Lab.

Inside Perlmutter is 6159 NVIDIA A100 Tensor Core GPUs, making it the largest A-100-powered system in the world. Researchers are waiting to use more than two dozen applications on the supercomputer that aim to advance science in astrophysics, climate science, and more. One of the projects that will be run on the supercomputer will help assemble the largest 3D map of the visible universe to date.

It will process data from the Dark Energy Spectroscopic Instrument, which is a type of cosmic camera able to capture as many as 5000 galaxies in a single exposure. The power of Perlmutter will help researchers capture dozens of exposures from one night to know where to point the instrument the next night for further study.

Scientists say that processing a year's worth of the data for publication would take weeks or months on prior systems, but using Perlmutter, the task can be accomplished in as little as a few days. Tensor cores of the type used inside the supercomputer can accelerate both double-precision floating-point math for simulations and the mixed-precision calculations required for deep learning. The machine supports important software, including OpenMP and other popular programming models supported by the NVIDIA HPC SDK.