Its calculation performance reaches 415.5 PFLOPS. Responsible for this data is the A64FX microprocessor created by Fujitsu and based on ARM 8.2A architecture. It also incorporates SVE (Scalable Vector Extensions) technology, to improve calculation results. This chip is the first ARM to support 512-bit SIMS SVE.
Each processor uses 32 GB of HBM2 memory with a bandwidth of 1TB / s. It works at 2.2 Ghz and the necessary packages have been added to complete the figure of almost 7.3 million cores and 5 PB of memory.
Fugaku is located at the Kobe RIKEN Computer Center in Japan. Its more than 150K nodes are placed there. They are interconnected through a high-speed Tofu Interconnect D network, which links ARMv8.2-A SVE chips of 52 cores per node.
It has a high bandwidth memory HBM2 with a capacity of 32 GiB per node. Its storage is 1.6 TB NVMe shared by each 16 nodes, to which 150 PB Shared FS are added, in addition to an extra cloud storage service.
As for the operating system governing this machine is Linux, specifically the RHEL 8 distribution and uses an IHK / McKernel simultaneously.
IBM’s well-known Summit supercomputer has been relegated to Second Place in this ranking, with 148.8 PFLOPS on HPL performance. The third position has been for Sierra, with 94.6 PFLOPS. The architecture of these last two computers is very similar, and they both include NVIDIA chips.
It should be noted that the tests carried out show an upward trend in both the aggregate performance of supercomputers and their energy efficiency.
For more information, see the Top500 official website.
0 Comments