HPE will build Japan’s most powerful AI supercomputer ABCI 3.0 based on NVIDIA H200

Japan’s National Institute of Advanced Industrial Science and Technology (AIST) has announced plans to build a new AI Bridging Cloud Infrastructure 3.0 (ABCI 3.0) supercomputer in Kashiwa, Chiba Prefecture, an update to the ABCI AI platform launched in 2018. The new supercomputer will be offered as a cloud service to both public and private organizations in the country, according to the NVIDIA blog.

The contractor is HPE, which will build a system using the Cray XD platform with NVIDIA H200 accelerators, combined with a 200G NVIDIA Quantum-2 InfiniBand interconnect. HPE did not disclose details about the total number of nodes, the cost of the system and the timing of its commissioning. According to The Register, we are talking about a system with 5U Cray XD670 nodes capable of accommodating eight NVIDIA H200/H100 accelerators and a pair of Intel Xeon Emerald Rapids. In addition, the ABCI-Q machine is being prepared based on NVIDIA H100 accelerators, focused on research in the field of quantum and hybrid computing.

Image source: AIST

HPE said ABCI 3.0 is expected to be the fastest AI supercomputer in Japan at approximately 6.2 Eflops (FP16?) or 410 Pflops (FP64). The ABCI 3.0 project is supported by Japan’s Ministry of Economy, Trade and Industry (METI) to strengthen the country’s computing resources through the Economic Security Fund. This is part of the broader $1 billion METI initiative, which includes both the ABCI program and investments in AI-powered cloud computing.