Intel and Argonne National Laboratory announced on the twenty second (local time) that they’ve accomplished the installation of the two exaflops (EFLOPS) class supercomputer ‘Aurora’ through a blog.
In March 2019, the U.S. Department of Energy and Intel decided to construct a supercomputer, Aurora, which is as much as 1,000 times faster than previous supercomputers by 2021.
Intel planned to finish Aurora by collaborating with HPE to equip the Cray supercomputer platform with the 4th generation Xeon scalable processor ‘Sapphire Rapids’ and data center GPU Max ‘Ponte Vecchio’, however the 4th generation Xeon scalable processor With delays in mass production of the GPU and data center GPU Max, full-fledged construction didn’t begin until November 2021. The development was accomplished in 1 yr and seven months.
The Aurora supercomputer consists of a complete of 10,624 blade servers. Each blade server is supplied with two Xeon CPU Max with high-bandwidth memory (HBM) and 6 data center GPU Max. Each blade server weighs roughly 31.75 kg.
Aurora has 21,248 CPUs and 63,744 GPUs, and is anticipated to be able to processing greater than 2 exaflops. Which means it might probably perform 200 quadrillion floating-point operations per second. The injected storage capability reaches 220 petabytes.
Currently, the fastest supercomputer on the planet is the AMD EPYC processor ‘Frontier’ installed on the US Department of Energy’s Oak Ridge National Laboratory. This supercomputer has a computational speed of as much as 1.194 exaflops. Since Frontier appeared early last yr, it has maintained its No. 1 position within the ‘Top 500,’ a worldwide supercomputer rating competition, for 18 months.
Nevertheless, when Aurora, which might operate 2 exaflops, is fully operational, it is anticipated to surpass the frontier and maintain its No. 1 position for a very long time.
“During acceptance testing, we will likely be using Aurora to coach some large-scale open source generative AI models for science,” said Rick Stephen, associate director of Argonne National Laboratory. The system is the right environment to coach these models.”
Reporter Park Chan cpark@aitimes.com