Atos has unveiled the XH300, its newest BullSequana supercomputer, which is significantly more powerful than the XH200, which powers the eighth fastest supercomputer in the world.
According to an Atos statement, the BullSequana XH300 integrates AMD, Intel, and Nvidia CPUs, GPUs, accelerators, and microprocessor technologies from French company The Silicon Pearl (SiPearl).
The BullSequana XH300 is described as ‘exascale-class’ by Atos, referring to the high-performance computing (HPC) industry’s target of benchmarking scores in the ‘exascale’ category.
Japan’s supercomputer tops the rankings
Most elite supercomputers have peak performance in the hundreds of Peta-Flop/s (equivalent to one quadrillion Flop/s). However, like Japan’s Fugaku, some have peak performance in the thousands of Peta-Flop/s or one Exa-Flop/s. However, this is under the ‘single or further-reduced precision’ operating mode.
Fugaku is the world’s fastest exascale supercomputer, according to Top500, which ranks the world’s fastest 500 supercomputers based on the High-Performance Computing Linpack (HPL) benchmark results in this mode of operation.
To be included in the Top500, supercomputer operators must submit their HPL scores.
Unseating Germany’s JUWEL Boost Module
According to Top500’s November 2021 list, Fugaku is the world’s fastest supercomputer. However, even with 7.6 million cores, it nevertheless hits a standard HPL benchmark of 442 Peta-Flop/s, and that’s still short of exascale but the highest of any supercomputer scores.
Fugaku’s scores are above those of the fastest US system, Summit, an IBM-built system at the Department of Energy’s Oak Ridge National Laboratory. It has 2.4 million cores. Summit has a 128 Pflop/s HPL score.
Europe’s current most powerful supercomputer is named the JUWEL Boost Module at Forschungszentrum Jülich in Germany. It has 449,280 cores and records 44 Pflop/s on the Linpack benchmark test.