Intel, HPE, and Argonne National Laboratory have announced at ISC High Performance 2024 that the Aurora Supercomputer has broken the Excascale barrier and is now the fastest AI supercomputer currently in existence.

Last November [1]Aurora debuted in second place on the TOP500 list as a partial deployment. But now the spring list is out for ISC 24 with an even better showing. The Aurora supercomputer is made up of 21,248 Intel Xeon CPU Max Series processors and 63,744 Intel Data Center GPU Max Series accelerators (GPUs). Aurora is spread across 166 racks with 10,624 HPE compute blades.

Aurora has placed second in the HPL LINPACK benchmark with 1.012 Exaflops while utilizing 9,234 nodes (87% of the total supercomputer). Aurora has placed third on the HPCG benchmark with 5,612 TF/s at 39% of the overall machine. So safe bet to say the fastest if fully utilized but seems the bring-up is still ongoing or at least at time of the benchmark submission. (Unfortunately the briefing around their Aurora ISC 2024 positioning was at the last minute on Sunday and collided with Mother's Day in the US to further complicate my usual work Sunday.) Aurora leads for AI at 10.6 Exaflops with HPL-MxP.

Intel's press release to hit the wire now goes on to tout the benefits of upcoming Intel Xeon 6 (Granite Rapids) and further out is the exciting Falcon Shores.



[1] https://www.phoronix.com/news/TOP500-H2-2023-List