Issue date: 2020.11.19
Release date: 2020.11.19
ABCI and Fugaku Achieve Highest Speeds in Machine Learning Processing Benchmark MLPerf HPC
At SC20, an international conference on HPC now online, Japanese supercomputers ABCI and Fugaku took first and second place in the MLPerf HPC, the benchmark for large-scale machine learning processing.
”ABCI” is the AI bridging cloud, a supercomputer system operated by the National Institute of Advanced Industrial Science and Technology (AIST). ”Fugaku” is a supercomputer under development by Riken and Fujitsu.
The MLPerf HPC is a benchmark for large-scale machine learning processes that require supercomputer-scale processing. The performance of two benchmark programs, CosmoFlow and DeepCAM, are competing against each other. CosmoFlow predicts cosmological parameters, and DeepCAM identifies extreme weather events.
Fujitsu, AIST, RIKEN, and Fujitsu Laboratories Ltd. will make their software stacks, including libraries and AI frameworks to accelerate large-scale machine learning processing, available to the public.