Japan’s National Institute of Advanced Industrial Science and Technology (AIST) is set to launch the ABCI 3.0 supercomputer in Kashiwa, Japan. This supercomputer will feature thousands of Nvidia H200 Tensor Core GPUs and is expected to be operational by the end of the year. The facility will offer 6 AI exaflops computing power and a general computing capability of 410 petaflops double precision. Each node will be connected via the Quantum-2 InfiniBand platform with high-speed bisectional bandwidth.
The HPE Cray XD system nodes will feature eight NVLink-connected H200 GPUs with high HBM3e memory capacity and speed, designed to accelerate generative AI and LLMs. The integration of advanced Nvidia Quantum-2 InfiniBand technology will ensure efficient communication for handling intensive AI workloads and large datasets.
ABCI 3.0 is a collaboration between AIST, HPE, and Nvidia, aiming to boost research and development capabilities for AI computing. The project has received support from Japan’s Ministry of Economy, Trade and Industry (METI) through the Economic Security Fund. Nvidia is working with METI on research and education initiatives to enhance Japan’s AI development capabilities.
The launch of ABCI 3.0 aligns with Nvidia CEO Jensen Huang’s vision of creating “AI factories” to handle large-scale AI tasks. This next-generation data center is seen as crucial for turning data into intelligence and driving modern economies worldwide. ABCI 3.0 will play a key role in promoting AI research and development in Japan, enhancing the country’s competitiveness in the field of AI.
Overall, the ABCI 3.0 supercomputer represents a significant advancement in Japan’s AI computing infrastructure, with a focus on accelerating AI research and development in collaboration with industry, academia, and government partners.
Article Source
https://insidehpc.com/2024/07/hpe-nvidia-6-ai-exaflops-supercomputer-to-be-installed-at-aist-research-organization-in-japan/