Intel supercomputer Aurora equipped with powerful storage and memory system

According to news on May 23, Aurora, the supercomputer designed by Intel for the Argonne National Laboratory in the United States, experienced a long delay, but was finally unveiled today.

According to ITBEAR technology information, Intel, HPE and Argonne are jointly developing a large-scale scientific computing language model called AuroraGPT. Compared with ChatGPT based on GPT-3, this model is built on a basic model with trillions of parameters and is larger in scale.

The Aurora supercomputer is powered by Intel's Xeon CPU Max and Xeon GPU Max series of chips and increases computing power from an initial 1 Exaflop to 2 Exaflop, making it comparable to the Frontier supercomputer based on AMD chips (ITBEAR Technology Information Note: Frontier is currently the most powerful supercomputer on the planet).

Aurora supercomputer contains a total of 10,624 nodes, including 21,248 Xeon CPUs based on the Sapphire Rapids-SP series and Ponte 63,744 GPUs designed by Vecchio. The peak bandwidth of the system reaches 2.12 PB/s, and the peak bisection bandwidth is 0.69 PB/s.

In terms of memory, the Aurora supercomputer is equipped with 10.9 PB of DDR5 DRAM, 1.36 PB of CPU HBM, and 8.16 PB of GPU HBM.

This means that the system’s DRAM has a peak bandwidth of 5.95 PB/s, the CPU HBM provides a peak bandwidth of 30.5 PB/s, and the GPU HBM reaches 208.9 Peak bandwidth of PB/s. In terms of storage, the system is equipped with 230 PB of DAOS capacity, runs at a peak bandwidth of 31 TB/s, and is configured with a total of 1,024 nodes.

Thanks to the blessing of Intel data center GPU Max series processors, Aurora has achieved more than NVIDIA A100 and AMD Instinct The fastest SimpleFOMP performance of the MI250X accelerator.

In addition, Intel also provides solutions that are not inferior to accelerators in terms of fusion reactor prediction, Monte Carlo methods (maximization) and QMCPACK (computing quantum mechanical properties).

The Aurora supercomputer will receive an upgrade later this year that will deliver peak performance of more than 2 Exaflops. It is worth mentioning that this supercomputer will also be equipped with the latest Aurora gen The AI ​​model provides computing power of one trillion parameters.

In addition to the Aurora supercomputer, Intel also announced a new data center GPU Max subsystem, which uses an x8 UBB design and is equipped with a total of 8 Ponte Vecchio GPU.

The above is the detailed content of Intel supercomputer Aurora equipped with powerful storage and memory system. For more information, please follow other related articles on the PHP Chinese website!