Intel has unveiled its highly anticipated Aurora supercomputer, designed for the Argonne National Laboratory. The system, which is set to deliver a massive performance boost, is powered by Intel’s Xeon CPU Max and Xeon GPU Max series. Originally planned to achieve 1 exaflop, the upgraded Aurora now aims to reach an impressive 2 exaflops, putting it on par with the world’s fastest supercomputer, Frontier, which is powered by AMD.
The specifications of the Aurora supercomputer have been revealed by Intel, showcasing its immense capabilities. It will consist of 10,624 nodes, including 21,248 Xeon CPUs based on the Sapphire Rapids-SP family and 63,744 GPUs based on the Ponte Vecchio design. With fabric interconnects boasting a peak injection bandwidth of 2.12PB/s and a peak bisection bandwidth of 0.69PB/s, the system ensures efficient data transfer.
Intel has collaborated with Argonne National Laboratory and HPE Cray EX to bring the Aurora supercomputer to life. The system will be built using an HPE Cray EX supercomputer, featuring 63,744 GPUs, 21,248 CPUs, and 1,024 DAUs. The memory includes DDR5 system DRAM of 10.9PB, CPU HBM capacity of 1.36PB, and GPU HBM capacity of 8.16PB. With 230PB of storage operating at a peak bandwidth of 31TB/s, the system is equipped with a total of 1,024 nodes.
One of the highlights of the Aurora supercomputer is its utilization of Intel’s Data Center GPU Max Series 1550, which outperforms NVIDIA A100 and AMD Instinct MI250X accelerators in terms of SimpleFOMP performance. It exhibits superior performance in fusion reactor prediction, the Monte Carlo method (maximization), and QMCPACK (calculation of quantum mechanical properties). Intel’s Xeon Max Series CPUs, the only x86 processors with high-bandwidth memory, demonstrate a 65% improvement over AMD’s Genoa processors in the HPCG benchmark.
In addition to the Aurora supercomputer, Intel introduces the data centre GPU Max Subsystem, featuring eight Ponte Vecchio GPUs in a compact x8 UBB design. This subsystem complements Intel’s data centre offerings and enhances the capabilities of AI and HPC workloads, providing researchers with increased efficiency and accuracy in tackling complex problems.
Intel’s commitment to advancing AI and HPC extends beyond hardware innovations. The company actively collaborates with research institutions, universities, and industry partners to drive advancements in software, algorithms, and system optimization. This comprehensive approach ensures that customers can harness the full potential of Intel’s hardware solutions, facilitating groundbreaking discoveries and insights.