Hewlett Packard Enterprise (HPE) has been selected to build an AI supercomputer for Japan's National Institute of Advanced Industrial Science and Technology (AIST).
The supercomputer, dubbed AI Bridging Cloud Infrastructure (ABCI) 3.0, will be built using HPE Cray XD systems featuring Nvidia H200 Tensor Core GPUs with each node interconnected by Nvidia Quantum-2 InfiniBand networking at 200GBps of bisectional bandwidth.
Each node will have eight NVLink-connected H200 GPUs with more than 140 gigabytes of HBM3e memory at 4.8 TBps.
In total, the system will have around 6.2 exaflops of half-precision performance, 410 double-precision petaflops, and is expected to be the fastest AI supercomputer in Japan.
The cluster will be available to both private and public organizations via a cloud service, and is housed in Kashiwa near Tokyo.
“Making AI accessible and inclusive is critical to bolstering innovation across industries and enhancing societal impact,” said Antonio Neri, president and CEO at HPE. “We are honored AIST selected HPE’s proven AI systems and expertise, combined with Nvidia's industry-leading accelerated computing platform, to power its AI cloud service. The AIST system will enable public and private organizations to advance research, drive innovation, and create value from generative AI.”
“In response to the escalating demand for generative AI and computational resources, AIST is maximizing the capabilities of ABCI 3.0 with unparalleled machine learning performance delivered through a powerful supercomputer,” said Dr. Yoshio Tanaka, executive officer, AIST. “We value our collaboration with HPE and Nvidia to deliver AI cloud services that rapidly train large-scale foundational models built from ever-growing datasets of image, acoustic, and sensor information, and we look forward to working more closely together to support the user community.”
AIST has been offering supercomputing services via the cloud since 2018. AIST’s ABCI systems have supported a range of scientific research, including the development of AI techniques to predict genome sequencing and materials informatics, and Japanese large language models.
It is the latest iteration of Japan’s large-scale Open AI Computing Infrastructure designed to advance AI R&D.
The ABCI 3.0 system is being supported by Japan's Ministry of Economy, Trade and Industry (METI) and is part of a broader $1 billion initiative by METI that includes both ABCI efforts and investments in cloud AI computing.
In June 2023, METI announced that it would be giving AIST $226m to develop a generative AI-focused supercomputer.
