Intel: Aurora supercomputer hits 2 exaflops of computing power, coming in 2022

Intel says its Aurora supercomputer for the Argonne National Laboratory will reach 2 exaflops of computing power. According to previous expectations, the supercomputer would offer 1 exaflops of computing power.

Intel announced on Wednesday that Aurora’s computing performance is two times higher than expected, writes SemiAnalysis Analyst Dylan Patel on Twitter. The supercomputer would deliver ‘more than two exaflops’ of computational performance. Aurora uses Intel’s upcoming Sapphire Rapids processors, along with Ponte Vecchio data center GPUs, and is intended for the US Department of Energy’s Argonne National Laboratory.

Aurora is scheduled to be delivered in 2022. Intel was initially expected to deliver the supercomputer in 2021, but the company has previously postponed this. Patel reports that it isn’t the first time Aurora has been delayed. Intel has also boosted Aurora’s computing performance several times. The supercomputer would initially become operational in 2018 with 180 petaflops, DataCenterDynamics writes. This was then postponed to 2021 by 1 exaflop, and the delivery was later pushed back to 2022 with 2 exaflops of computing power.

In the meantime, the Argonne National Laboratory is taking delivery of another Polaris supercomputer, which uses 560 AMD EPYC processors with 32 cores, in addition to 2240 Nvidia A100 accelerators. According to the Argonne National Laboratory, Polaris achieves theoretical computing power of up to 1.4 exaflops in AI computing tasks. Polaris will be used to prepare software for Intel’s Aurora supercomputer. Polaris will be delivered and installed this year and should be operational by early 2022.