AMD and Nvidia join forces to create Polaris supercomputing
Intel won the bid for the supercomputer project of the U.S. Department of Energy Argonne National Laboratory in 2019, Aurora supercomputing system will be built based on Sapphire Rapids and Ponte Vecchio to achieve ExaFLOP computing power. However, Intel has been delayed for several months due to R&D issues, the Aurora supercomputing system may not be available until 2022 to 2023. The US Department of Energy has decided to purchase a supercomputer built with AMD and NVIDIA chips. The supercomputing system is called Polaris.
The Polaris supercomputing system will be launched this year as a software test machine of the Argonne National Laboratory and will do the preliminary work for the future Aurora supercomputing system. Nvidia officially stated that the Polaris supercomputing system will use four Nvidia A100 computing cards and two AMD EPYC 7532 processors on each node. There will be a total of 560 nodes, providing 44 PetaFLOPS of FP64 performance. Although it is weaker than the Aurora supercomputing system, it can still enter the top ten of the global supercomputing Top500 list.
The Polaris supercomputer system supports Cray Slingshot high-performance scalable interconnection architecture, which means that this supercomputer is built by HPE. It should be similar to the Perlmutter used by the Lawrence Berkeley National Laboratory of the National Energy Research Scientific Computing Center. By March 2022, the Polaris supercomputing system will be upgraded, from Slingshot 10 to Slingshot 11 architecture to match the Aurora supercomputing system, and will be upgraded to EPYC 7543 processor.