Nvidia GPUs will power Argonne National Laboratory’s Polaris supercomputer, which is being built by Hewlett Packard Enterprise, and will be available to early-stage users early next year to tackle a variety of research tasks.
Described by Nvidia as the largest GPU-based supercomputer at Argonne, Polaris will have 560 total nodes, each with four Nvidia A100 GPUs, translating to a total of 2,240 A100 Tensor Core GPUs. The system will be able to achieve almost 1.4 exaflops of theoretical AI performance and approximately 44 petaflops of peak double-precision performance.
“We’re now in the era of exascale performance for supercomputing,” said Dion Harris, head of data center product marketing at Nvidia, adding that Nvidia is helping to address the challenges posed by the end of Moore’s Law, a time during which computing tasks have continued growing.
“Argonne is on a path to exascale computing, and we’ll deliver over an exaflop of performance for AI computing and acceleration of their core applications,” Harris said.
Polaris will be hosted at the Argonne Leadership Computing Facility. Argonne, a U.S. Department of Energy lab, is located in Lemont, Illinois, about 30 miles southwest of Chicago.
“Polaris is a powerful platform that will allow our users to enter the era of exascale AI,” said ALCF Director Michael E. Papka, in a statement. “Harnessing the huge number of NVIDIA A100 GPUs will have an immediate impact on our data-intensive and AI HPC workloads, allowing Polaris to tackle some of the world’s most complex scientific problems.”
According to Argonne, Polaris will accelerate transformative scientific exploration, such as advancing cancer treatments, exploring clean energy and propelling particle collision research to discover new approaches to physics. Polaris will also be available to researchers from academia, government agencies and industry through the ALCF’s peer-reviewed allocation and application programs.
Though Polaris is not online yet, it’s already at the center of some controversy. Nvidia’s Polaris announcement hit the newswires just as Reuters reported that Argonne turned to Nvidia and AMD to help create a supercomputer amid delays in the construction of another supercomputer on site, the Intel-based Aurora. That project was announced in 2019.
None of the parties mentioned in the story commented to Reuters. Nvidia, for its part, did not mention AMD during a media presentation about Polaris earlier this week. Regarding Aurora, Harris made clear that Polaris would come online separately from Aurora, but eventually could coordinate with Aurora.
Nvidia’s press release announcing Polaris also said the system will allow researchers “to update their scientific workloads for Aurora, Argonne’s forthcoming exascale system.”
RELATED: Intel unveils exascale Ponte Vecchio GPU, part of Aurora