Department of Energy and HPE unveil Polaris supercomputer for advanced AI and HPC research.
A delay in the delivery of Intel Corporation's exascale supercomputer to the U.S. Department of Energy's Argonne National Laboratory has created a new opportunity for chipmaker rivals NVIDIA Corporation and Advanced Micro Devices (AMD). On Wednesday, Argonne unveiled Polaris, a testbed supercomputer made up of 280 of Hewlett Packard Enterprise's Apollo Gen10 Plus systems and powered by 2,240 of NVIDIA's A100 Tensor Core graphics processing units and 560 of AMD's second- and third-generation EPYC processors. Polaris is expected to deliver approximately 44 petaflops of peak double precision performance and almost 1.4 exaflops of theoretical artificial intelligence performance, up to four times faster performance than Argonne's current supercomputers.
The machine will enable scientists and developers to test and optimize software codes and applications that will be used by Intel's $500 million Aurora supercomputer for a range of scientific projects, from advancing cancer treatments and the nation's energy security to expanding the boundaries of physics with particle collision research. "Polaris is well equipped to help move the [Argonne Leadership Computing Facility] into the exascale era of computational science by accelerating the application of AI capabilities to the growing data and simulation demands of our users," ALCF director Michael Papka said in a statement. Polaris, which was delivered and installed in August, will go into use starting in early 2022 before opening to the broader high performance computing community in the spring of 2022.
Aurora was originally supposed to be delivered in 2021, but has been delayed for several months. The delay is being caused by Intel's Sapphire Rapids core processing units and Ponte Vecchio graphics processing units, key components of the Aurora system. An Intel spokesperson confirmed to FOX Business that Aurora remains on track for delivery next year, with the chips expected to go into production in 2022. A spokesperson for Argonne National Laboratory confirmed that the Aurora system will be faster than Polaris, providing greater than 1 exaFLOP of double precision peak performance.
The struggle for Intel presents an opportunity for NVIDIA and AMD, which are battling alongside Intel for dominance in the data center chip market. AMD will help power the National Nuclear Security Administration's $600 million El Capitan supercomputer, which is set to be delivered to Lawrence Livermore National Laboratory in early 2023, and is expected to be the world's fastest exascale computer, producing record-breaking speed of 2 exaflops, 10 times faster than the current most powerful supercomputer. Meanwhile, NVIDIA is currently powering Oakridge National Laboratory's Summit supercomputer, developed in collaboration with IBM, which can produce 200 petaflops. According to Technavio, a global technology research and advisory company, the data-center chip market has the potential to grow by $18.77 billion between 2021 and 2025.