The Aurora supercomputer "breaks the exascale barrier" with 1,012 exaflops

May 13. (Portaltic/EP) –

dawn has become the fastest supercomputer dedicated to artificial intelligence (AI) for open science in the world, after surpassing the exascale barrier with 1,012 exaflops, with 87 percent of the system.

The Aurora supercomputer is part of a Laboratory project Argonne National (ANL) of the United States and technology firms Intel and Hewlett Packard Enterprise (HPE) to advance science and engineering powered by AI.

Among the works carried out with this team are the training of a generative AI ‘chatbot’, called Aurora GPTand expanding capabilities with exascale computingthat is, with the capacity to perform a minimum of one exaflop operation per second.

Within the framework of ISC High Performance 2024, held in Hamburg (Germany), Intel has announced that Aurora has “broken the exascale barrier” by reaching 1,012 exaflops, for which it has used 9,234 nodeswhich they suppose “only” 87 percent of the system.

This means that Aura has become the world’s fastest AI system dedicated to AI for open use in science, achieving 10.6 exaflops of artificial intelligence, as Intel reports in a Press release.

This supercomputer has also achieved second place in the High Performance LINPACK (HPL) benchmark and third in the High Performance Conjugate Gradient (HPCG) benchmark at 5,612 teraflops per second (TF/s) with the 39 percent of the system.

Aurora is comprised of an expansive system with 166 racks, 10,624 compute blades, 21,248 Intel Xeon CPU Max Series processors and 63,744 Intel Data Center GPU Max Series units. Also with open Ethernet-based supercomputing interconnection in a single system of 84,992 HPE Slingshot Fabric endpoints.