The supercomputer, named after Nobel Prize-winning astrophysicist Sol Perlmutter, will initially be tasked with creating the most accurate maps of the universe ever, using more than 6000 NVIDIA GPUs. A100 Tensor Core.
In its current state, Perlmutter is capable of delivering approximately 4 exaflops exaFLOPS (computer performance unit) of AI performance, which Nvidia asserts makes it “the fastest system on the planet using the 16- and 32-bit mixed math used by AI”. Its performance is set to be further enhanced as part of “Phase Two”, with the introduction of a second chip of CPU cores.
Once the system is complete, it is expected to be included in the top 5 supercomputers in the Top 500 rankings.
According to Nvidia, more than 7,000 researchers will operate the Perlmutter system with the goal of advancing astrophysics and climate science..
In one project, a supercomputer will help assemble the largest 3D map of the visible universe to date. It will process data from a dark energy spectrophotometer(DESI), which is a type of cosmic camera that can capture up to 5,000 galaxies in a single exposure, and researchers need the speed of the Perlmutter GPUs to capture dozens of exposures from one night to know where to point the camera DESI the next night. Preparing data collected over a year could take weeks or months on earlier systems, but Perlmutter should help them get the job done in a few days..
Nvidia hopes that, once complete, the map of the universe will help researchers learn more about dark energy, the force behind the accelerating expansion of the universe. The discovery of dark energy in 2011 led to Sol Perlmutter receiving the Nobel Prize.
This work is likely to be useful in the field of materials science, by potentially revealing ways to develop more efficient batteries, biofuels, and the like.