New AI supercomputer will help create the largest-ever 3D map of the universe

Perlmutter could complete tasks in days that would normally take months.


The newest big-name supercomputer might help solve some of astrophysics' most important questions. VentureBeat reports the National Energy Research Scientific Computing Center has officially dedicated Perlmutter, billed as one of the fastest supercomputers for AI, and it will start by helping to build the largest-ever 3D map of the visible universe to study the dark energy accelerating the cosmos' expansion.

The machine will process data from the Dark Energy Spectroscopic Instrument to guide observations. Perlmutter will process "dozens" of galaxy exposures from a given night's study to dictate where DESI aims next.

As AMD notes, the initial version of Perlmutter includes 1,536 nodes that each have a 64-core Epyc 7763 processor and four NVIDIA A100 GPUs. The result is a system with almost four exaflops of performance (plus 35 petabytes of storage) when handling AI tasks, and dramatically reduced calculation times. Perlmutter could process a year's worth of DESI info in a "few days" instead of weeks or months, NVIDIA claimed.

A second phase due later in 2021 will add 3,072 CPU-only nodes that each have dual Epyc 7763 chips.

The map isn't the only project on deck. Perlmutter will also study atomic interactions to foster clean energy tech like biofuels, for example. The universe map is ambitious, though, and helps illustrate just how far supercomputers have come in tackling truly large-scale work.