Perlmutter Is The World's Fastest AI Supercomputer with 6,144 Nvidia GPUs

On May 27, 2021, tech giant Nvidia and the National Free energy Inquiry Scientific Computing (NERSC) switched on, what they phone call, the globe'due south fastest AI supercomputer. Named after Saul Perlmutter, an astrophysicist at the Lawrence Berkley National Laboratory, the "Perlmutter" is hailed to exist the fastest supercomputer for conducting artificial intelligence (AI)-related tasks.

Nvidia'southward production marketing atomic number 82 revealed details of the Perlmutter via an official blog post. Harris stated that Perlmutter "is the fastest arrangement on the planet" for processing workloads including 16-bit and 32-chip mixed-precision math that is used in diverse bogus intelligence applications.

However, initially, Perlmutter will be tasked to create the largest-e'er 3D map of our universe before moving on to other AI-based projects in the future. Moreover, the researchers will add fifty-fifty more AI computing power to the system later this year, under the second phase of its evolution.

"In ane project, the supercomputer volition help gather the largest 3D map of the visible universe to appointment. It will process data from the Night Energy Spectroscopic Instrument (DESI), a kind of cosmic camera that tin can capture as many equally v,000 galaxies in a single exposure," wrote Dion Harris in the weblog postal service.

Now, coming to the internals of the computers, the Perlmutter packs 6,144 Nvidia A100 Tensor Core GPUs. Well, permit that sink in for a moment!

And so, over vi yard Nvidia GPUs will power the organization to process some of the most complex AI-based tasks. It is expected to give over seven,000 NERSC researchers access to four exaflops of mixed-precision computing operation for AI-based scientific researches.

"Traditional supercomputers can barely handle the math required to generate simulations of a few atoms over a few nanoseconds with programs such as Breakthrough Espresso. But by combining their highly authentic simulations with machine learning, scientists can written report more atoms over longer stretches of fourth dimension," said Wahid Bhimji, the acting caput of NERSC's data and analytics services team.