A Closer Look at a Deep Learning Library for the Fugaku A64FX CPU

Kentaro Kawakami shares the development story behind getting Intel® oneAPI Deep Neural Network Library (oneDNN) on Arm* architecture for the Fugaku A64FX CPU. Fujitsu managed to make full use of Arm SVE architecture, and succeeded in improving performance by 9.2 times in training and 7.8 times in inference. Using oneDNN, Fujitsu managed to achieve the best performance as a CPU with MLPerf HPC v0.7.

Kawakami and his team optimized and ported the oneDNN library (which continues to be developed as open-source software) for the Arm v8-A instruction set so that it can run at high speed on the Fugaku supercomputer. The new Fugaku supercomputer has been delivered to Port Island located off the coast of Kobe, Japan. Developed jointly by RIKEN and Fujitsu, this supercomputer has entered the trial run phase. As of June 2020, it had already won four "firsts" in worldwide supercomputer rankings (TOP500, HPCG, HPL-AI, Graph500), so it is off to a very promising start.

 

Speaker

Kentaro Kawakami is the senior researcher at the Platform Innovation project, Fujitsu Laboratories Ltd. He joined Fujitsu Laboratories in 2007. He has been involved in R&D of image codec LSIs and wireless sensor nodes, and is currently engaged in R&D of AI software for Arm HPC. His department researches and develops techniques to accelerate deep learning processes on Fugaku, PRIMEHPC FX1000/700, and GPU-based supercomputers. His GitHub* account name is "kawakami-k". Kawakami-san lives in Japan and loves cats.