Over 63K Xeon GPU Max and 21K Xeon CPU Max Chips

Intel & Argonne National Laboratory announces the successful blade installation in the Aurora supercomputer, bringing it one step closer to full functionality.

Intel-based Aurora supercomputer has 2 ExaFLOPS of computing power, potentially surpassing AMD’s limit

The Aurora supercomputer has been subject to several delays since its inception, but we can finally see it running. For those unaware, the Aurora supercomputer features Intel’s Xeon CPU Max and Xeon GPU Max series, boosting performance to 2 ExaFLOPS. One of the applications of the Aurora platform will be to provide a modern generative AI model for science.

It offers 10,624 nodes with 21,248 Xeon CPUs from the Sapphire-Rapid SP series. It comes with a total of 63,744 GPUs based on the Ponte Vecchio design, which allows it to offer a peak injection of 2.1[ads1]2 PB/s and a peak halving bandwidth of 0.69 PB/s.

Here’s how the Intel-powered Aurora supercomputer has an advantage, as described by Intel Super Compute Group VP Jeff McVeigh earlier:

• Intel Data Center GPU Max Series outperforms Nvidia H100 PCIe cards by an average of 30% on various workloads1, while independent software vendor Ansys shows a 50% speedup for Max Series GPUs over H100 on AI-accelerated HPC applications.
• The Xeon Max Series CPU, the only x86 processor with high-bandwidth memory, shows a 65% improvement over AMD’s Genoa processor on the High Performance Conjugate Gradients (HPCG) benchmark1, and uses less power. High memory bandwidth has been noted as one of the most desired features for HPC customers.
• 4th Gen Intel Xeon Scalable processors – the most widely used in HPC – deliver an average speedup of 50% over AMD’s Milan4, and energy company BP’s latest 4th Gen Xeon HPC cluster delivers an 8x increase in performance over the previous generation processors with improved energy efficiency.
• The Gaudi2 deep learning accelerator performs competitively in deep learning training and inference, with up to 2.4x faster performance than the Nvidia A100.

For memory capacity, the Aurora supercomputer has 10.9 PB of DDR5 system DRAM, 1.36 PB of HBM capacity through the CPUs, and 8.16 PB of HBM capacity through the GPUs. Also, it uses an arrangement of 1024 storage nodes which gives a total capacity of 220TB. If you’re curious about how this gigantic system will be utilized, here’s a quick explanation:

From tackling climate change to finding cures for deadly diseases, scientists face monumental challenges that require advanced computing technologies at scale. Aurora is poised to meet the needs of the HPC and AI communities, providing the necessary tools to push the boundaries of scientific exploration.

The latest Intel Data Center GPU Max Series 1550, operating on Aurora, provides the best SimpleFOMP performance, beating the NVIDIA A100 and AMD Instinct MI250X accelerators. However, the supercomputer has yet to pass preliminary testing. After that, it is expected to appear on the Top500.org list, potentially overtaking the AMD-powered Frontier supercomputer. The Aurora supercomputer is on track to be fully functional by this year.