Nikkei in Japan reports that Fujitsu is building a 37 Petaflop supercomputer for the National Institute of Advanced Industrial Science and Technology (AIST).
Targeted at Deep Learning workloads, the machine will power the AI research center at the University of Tokyo’s Chiba Prefecture campus. The new Fujitsu system feature will comprise 1,088 servers, 2,176 Intel Xeon processors, and 4,352 NVIDIA GPUs.
The supercomputer will have a theoretical processing power of 37 petaflops (1 petaflops equals 1 quadrillion calculations per second). That is over three times more powerful than the K computer at Riken, another government-affiliated research institute, currently No. 1 in Japan. The hardware will be among the fastest supercomputers devoted to AI in the world.
To build the machine, Fujitsu will use 1,088 servers, 2,176 Intel central processing units and 4,352 Nvidia graphics processing units. The company will achieve ultrafast processing speeds by applying its technology for efficient data transmissions between large numbers of chips. Fujitsu has also developed software to make AI run smoothly.
Back in May, Fujitsu previewed quantum-inspired technology in the field of artificial intelligence, focusing on the areas of combinatorial optimization and machine learning. The companies will work together in both the Japanese and global markets to develop applications which address industry problems using AI developed for use with quantum computers. This collaboration will enable software developed by 1QBit for quantum computers to run on a “digital annealer,” jointly developed by Fujitsu Laboratories Ltd. and the University of Toronto. A digital annealer is a computing architecture that can rapidly solve combinatorial optimization problems using existing semiconductor technology.