NVIDIA has officially unveiled a new generation GPU architecture with a 7nm process, Ampere, featuring high-performance computing, which soon came to fruition. Atos has announced the latest generation of supercomputing system, the BullSequana X2415, which integrates the A100 GPU of NVIDIA ampere architecture for the first time, alongside the AMD Dragon processor, which is also a 7nm process.
The new server is based on the NVIDIA HGX-A100 module, each with four NVIDIA A100 GPUs, interconnected via the third generation NVLink bus, paired with two AMD EPYC Dragon processors and four NVIDIA Mellanox InfiniBand network ports.
The model of AMD Dragon processor is not disclosed, not surprisingly should be customized high-performance version of the Dragon 7H12, 64 core 128 thread, the reference frequency increased to 2.6GHz, the maximum acceleration frequency 3.3GHz, the three-stage cache 256MB, thermal design power consumption 280W.
The JUWELS BullSequana supercomputing system at the Ulich Research Center in Germany will be the first to be deployed and is expected to be available online this summer, with peak computing capacity of 70PFlops, making it the strongest supercomputing system in Europe.
According to the November 2019 Global Supercomputing TOP500 list, this calculation is also ranked fifth in the world, while the second-ranked U.S. Sierra Supercomputing uses the NVIDIA GV100 GPU.