Intel Unveils Aurora Supercomputer and Breakthrough genAI Model for Scientific Advancements

Intel has announced the development of the Aurora supercomputer, consisting of 21,248 Xeon CPUs and 63,744 GPUs, with a computational power exceeding 2 ExaFlops, in collaboration with the Argonne National Laboratory.

According to reports, the Aurora supercomputer, with its impressive 2 ExaFlops computational power, is a computing monster. It will employ the Aurora genAI model, which will be trained on general and scientific content, scientific data, and related materials. Aurora genAI is expected to be a cutting-edge scientific application model, driving advancements in fields such as systems biology, cancer research, atmospheric science, astronomy, polymer chemistry, and materials science.

Let's delve into the details of the Aurora supercomputer. Intel revealed that the Aurora supercomputer will consist of a total of 10,624 nodes, including 21,248 Xeon CPUs from the Sapphire Rapids-SP series, and 63,744 GPUs based on the Ponte Vecchio design. The system will feature a highly interconnected architecture, boasting a peak input bandwidth of 2.12PB/s and a peak load-balancing bandwidth of 0.69PB/s.

In terms of memory capacity, the Aurora supercomputer will be equipped with 10.9PB of DDR5 system DRAM, 1.36PB of CPU HBM, and 8.16PB of GPU HBM. The system DRAM will offer a peak bandwidth of 5.95PB/s, while the CPU HBM will reach 30.5PB/s and the GPU HBM an impressive 208.9PB/s. In terms of storage, the supercomputer will provide a capacity of 230PB with a peak bandwidth execution of 31TB/s using DAOS technology.

Leveraging the powerful computing capabilities of the Aurora supercomputer, Intel aims to develop the Aurora genAI model based on Megatron and DeepSpeed, with a target of generating 1 trillion parameters. This is a significant increase compared to the free and public versions of ChatGPT, which have a parameter target of only 175 million. By generating such a large number of parameters through Aurora genAI, Intel hopes to provide comprehensive knowledge from millions of sources, offering valuable insights and directions for the development of pure scientific applications.

This groundbreaking announcement opens up new opportunities for the electronic components industry to cater to the growing demand for high-performance computing solutions in the scientific community. The Aurora supercomputer and the Aurora genAI model represent a significant leap forward in computational capabilities, enabling advancements in scientific research and discovery across various disciplines.