The Aurora Supercomputer is Ready with Intel Max Series CPUs and GPUs Under the Hood

The Aurora supercomputer will be coming online later this year.

(Image credit: Argonne National Laboratory)

On Thursday, 22nd July 2023, Argonne National Laboratory and Intel announced that installation of the 10624 blades making up the Aurora supercomputer have been successfully installed with the supercomputer slated to come online later in 2023.

The machine uses tens of thousands of Xeon Max ‘Sapphire rapids’ processors with the highest bandwidth memory on the planet – HBM2E (high bandwidth memory 2). It also boasts thousands of Data Centre GPU Max ‘Ponte Vecchio’ compute GPUs for an overall performance of over 2 FP64 Exaflops (1 Exaflop measures that a system can calculate at least one quintillion floating-point operations per second). To put that into perspective, the RTX 3060 will process 199 Gigaflops per second – 0.00001% of the Aroura’s performance.

Aurora is comprised of 166 racks with 64 blades per rack, totalling 10624 blades. Each blade is based on two Exon Max CPUs with 64GB on-package HBM2E memory as well as six Intel Data centre Max ‘Ponte Vecchio’ compute GPUs. Each CPU and GPU is cooled by a custom liquid-cooling system.

Overall, Aurora packs an impressive 21248 general purpose CPUs with over 1.1 million high performance cores, 19.9PB (petabytes or 19900000 Gigabytes) of DDR5 memory, 1.36PB of HBM2E memory directly attached to the CPUs, and 63744 compute GPUs designed for massively parallel AI and HPC (high performance computing) workloads with 8.16PB of HBM2E memory onboard. The blades are interconnected using HPE’s Slingshot fabric designed for supercomputers.

Layout of Aroura blades. (Image credit: Intel)

While the instillation of the Aurora blades has been completed, the supercomputer is yet to pass acceptance testing. Should testing go well, later in the year Aurora will come online, being the first supercomputer to achieve 2 Exaflops of performance.

“While we work toward acceptance testing, we are going to be using Aurora to train some large-scale open-source generative AI models for science,” said Rick Stevens, Argonne National Laboratory associate laboratory director. “Aurora, with over 60,000 Intel Max GPUs, a very fast I/O system, and an all-solid-state mass storage system, is the perfect environment to train these models.”

According to intel, preliminary tests with the Max series GPUs show they excel in real-world science and engineering workloads, delivering twice the performance of competing options. In addition, Intel says its Intel Exon Max series CPUs offer a 40% advantage in performance of its rivals in numerous real world HPC applications.

Similar Posts