ACADEMIA
Japan Advanced Institute of Science and Technology Selects SGI to Advance High-Speed Simulations and Next Generation AI Research
SGI announced that the Japan Advanced Institute of Science and Technology (JAIST), a leading university in Nomi, Ishikawa Prefecture, headed by Dr. Takuya Katayama, has selected an SGI UV 1000 high performance computing (HPC) solution for its next-generation science and technology simulation system. This SGI UV 1000, the most powerful model in the SGI UV series, features 192 sockets (1,536 cores) powered by Intel Xeon processor E7 family series plus 12TB of shared memory, and has begun operation at the Institute.
Since its establishment in April 1991, JAIST’s Research Center for Advanced Computing Infrastructure, under the direction of Koichiro Ochimizu, has deployed a wide variety of computer systems to support advancements in cutting-edge scientific and technological fields and respond to fast-paced changes. These resources have been provided to all of JAIST’s faculty and students. In recent years, system users have developed a need for faster, larger-scale simulations and the center responded to this need by upgrading its core information system to an SGI UV 1000 HPC solution, making it the largest single-system shared memory computer in Japan.
JAIST plans to utilize the system for the following purposes:
-- Solid-state analysis and new materials development using high-speed simulations at molecular and atomic levels
-- Development of medical diagnostic and treatment support systems using simulations of the flow of particles within the body
-- Development of encryption technology for personal data protection and inference systems for artificial intelligence research
The SGI UV 1000 system provides two distinct advantages for JAIST. First, the system features a single, massive shared memory, which allows large-scale scientific and technological simulation programs to run without limits on memory usage. Second, the single Linux image facilitates sophisticated parallel processing, which is convenient for educational research.
“The increasing complexity of challenges that organizations such as JAIST are working to solve is the reason why we remain committed to developing and providing solutions that serve both industry and education,” said Ryutaro Ishimoto, president of SGI Japan. “With this new HPC solution now in service to the Institute, the results they achieve will strengthen our understanding of a wide range of critical problems and lay the groundwork for new discoveries now and well into the future.”
The SGI UV 1000 can scale to a maximum of 256 sockets (2,560 cores) and 16TB in single system image running Linux. All of the system memory can be shared, enabling optimal use of memory when running large-scale simulations and alleviating any programming restrictions. The JAIST system parallel computer with a cache-coherent shared memory architecture comprises of 96 processor memory blades, each configured with two Intel Xeon E7 series processors (16-cores) and 128GB of memory, using a NUMAlink 5 (15GB/second) interconnect.