Mellanox InfiniBand to Accelerate Japan’s Fastest Supercomputer for AI Applications


Mellanox Technologies, Ltd., a global supplier of high-performance, end-to-end smart interconnect solutions for data center servers and storage systems, has announced that Fujitsu has selected InfiniBand solutions for the National Institute of Advanced Industrial Science and Technology (AIST)’s the AI Bridging Cloud Infrastructure (ABCI) supercomputer, the fastest supercomputers in Japan, and world’s first large-scale open artificial intelligence Infrastructure.

The ABCI will be deployed by Fujitsu at the University of Tokyo and is scheduled to begin operations in the second quarter of 2018. The dual EDR 100G InfiniBand network, will provide an aggregated data throughput of 200Gb/s, connecting the Fujitsu PRIMERGY servers, including multiple NVIDIA GPUs and NVMe storage.

The Fujitsu system is expected to deliver a theoretical peak performance of 550 petaflops in half-precision floating point operations, or 37 petaflops of double-precision floating point operation performance.

“We are proud to work with AIST to bring the world-leading interconnect technology and solution to accelerate the fastest supercomputer in Japan,” said Gilad Shainer, vice president of marketing at Mellanox Technologies. “The key HPC and AI interconnect accelerators we offer, such as SHARP, GPUDirect, RDMA and NVMe over Fabrics, deliver the highest performance, efficiency and scalability to HPC and AI applications.”

“We are happy to collaborate with Mellanox and leverage the smart InfiniBand offload engines to deliver the best performance for artificial intelligence applications,” said Yujiro Maki, head of data center server division at Fujitsu, Japan. “The combination of Fujitsu PRIMERGY Server technology and Mellanox interconnect solution will enable AIST to meet their research goals. We look forward to continue working together with Mellanox for future opportunities.”