Nvidia, NERSC claim Perlmutter is world’s quickest AI supercomputer

Nvidia and the National Vitality Be taught Scientific Computing Center (NERSC) on Thursday flipped the “on” switch for Perlmutter, billed because the world’s quickest supercomputer for AI workloads. Named for astrophysicist Saul Perlmutter, the fresh supercomputer boasts 6,144 NVIDIA A100 Tensor Core GPUs and should quiet be tasked with stitching collectively a truly distinguished ever 3D plot of the seen universe, amongst a form of initiatives.

Perlmutter is “the quickest system on the planet” at processing workloads with the 16-bit and 32-bit combined-precision math aged in synthetic intelligence (AI) functions, stated Nvidia global HPC/AI product marketing lead Dion Harris someday of a press briefing earlier this week. Later this year, a 2nd allotment will add even extra AI supercomputing power to Perlmutter, which is housed at NERSC on the Lawrence Berkeley National Laboratory.

“In a single project, the supercomputer will aid assemble a truly distinguished 3D plot of the seen universe to this point. This can direction of data from the Sad Vitality Spectroscopic Instrument (DESI), a form of cosmic camera that can well perchance opt as many as 5,000 galaxies in a single exposure,” Harris wrote in a weblog post announcing the news.

“Researchers need the charge of Perlmutter’s GPUs to opt dozens of exposures from one night to clutch where to point DESI the next night. Making ready a year’s value of the information for e-newsletter would opt weeks or months on prior techniques, nonetheless Perlmutter should quiet aid them carry out the project in as minute as about a days,” he wrote.

Supercharging HPC with AI and machine studying

Firing up an AI-optimized supercomputer “represents a extraordinarily staunch milestone,” stated Wahid Bhimji, acting lead for NERSC’s data and analytics companies crew.

“AI for science is a declare spot on the U.S. Department of Vitality, where proof of ideas are transferring into manufacturing employ circumstances in areas worship particle physics, provides science, and bioenergy,” he stated.

“Of us are exploring greater and greater neural-community devices and there’s a quiz for procure entry to to extra extremely efficient resources, so Perlmutter with its A100 GPUs, all-flash file system, and streaming data capabilities is well timed to meet this need for AI,” Bhimji added.

Perlmutter will give NERSC’s roughly 7,000 supported researchers procure entry to to four exaflops of combined-precision computing performance for AI-assisted scientific initiatives. Moreover the DESI mapping project, researchers are teeing up time with the supercomputer for work in fields worship local climate science, where Perlmutter will aid in probing subatomic interactions to behold inexperienced power sources.

That project, which is in a plan to generate simulations of atoms interacting, requires the special mix of AI and excessive-performance computing (HPC) that Perlmutter delivers, Harris stated.

“Worn supercomputers can barely take care of the arithmetic required to generate simulations of some atoms over about a nanoseconds with functions fair like Quantum Espresso. But by combining their extremely ethical simulations with machine studying, scientists can ogle extra atoms over longer stretches of time,” he stated.

The skill to leverage AI in supercomputing additionally has researchers optimistic in regards to the DESI project. Moreover mapping the known universe, the project “targets to shed gentle on darkish power, the mysterious physics unhurried the accelerating growth of the universe,” NERSC data architect Rollin Thomas stated. System namesake Saul Perlmutter, who remains a working astrophysicist at Berkeley Lab, was awarded the 2011 Nobel Prize for Physics for his contributions to the invention of darkish power.

“To me, Saul is an instance of what other americans can carry out with the elegant mixture of insatiable curiosity and a dedication to optimism,” Thomas stated.

He added that in preparatory work with researchers to procure code ready for Perlmutter supercomputer workloads, NERSC was already seeing 20x sooner GPU processing performance than in previously accessible techniques.