The system’s total theoretical processing performance will reach 4 petaflops. The system will be comprised of two server architectures, with 24 of NVIDIA® DGX-1™ servers and 32 FUJITSU Server PRIMERGY RX2530 M2 servers, along with a high-reliability, high-performance storage system.
Fujitsu is leveraging the extensive know-how that it and Fujitsu Laboratories Ltd. have in high-performance computing development and AI research to build and operate one of Japan’s most advanced AI research systems. The company will also provide support for R&D that utilizes the system, thereby contributing to the creation of a future society in which AI is used to find solutions to a variety of social issues.
About the Deep learning system
The new system will be used at the Center for Advanced Intelligence Project to accelerate R&D into base technologies for innovative AI and the development of technologies that work to support such fields as regenerative medicine and manufacturing, and that into the future enable real-world implementation of solutions to social issues, including healthcare for the elderly, management of aging infrastructure, and response to natural disasters.
The Center for Advanced Intelligence Project, which has an integrated R&D system for everything from basic research to public implementation, advances joint research with researchers in a variety of universities, research institutes, clinical medical organizations, and in the world of industry. The new system will support AI researchers in Japan, and will become a core system that spurs on breathtaking advances in research that realizes innovative AI for the world.
Overview of the Deep learning system
The system is comprised of two server architectures specialized for deep learning using the latest CPUs and GPUs, and a storage system; it is being installed in Fujitsu’s Yokohama datacenter, a robust facility with cutting-edge security. Along with the standard DGX-1 deep learning software environment which NVIDIA provides in a public cloud, Fujitsu integrated a customized software environment for use in a secure on-site network. The system has operations management functions for easily and flexibly creating and reproducing calculation execution environments and the security and reliability for processing data of high importance, such as personal and intellectual property data.
Configuration of Deep learning system
1. Computation server
With 24 NVIDIA DGX-1 servers, each including eight of the latest NVIDIA® Tesla® P100 accelerators and integrated deep learning software, and 32 FUJITSU Server PRIMERGY RX2530 M2 servers, the system has a total theoretical performance of more than 4 petaflops (when performing half-precision floating-point calculations).
In building the system, an early deployment and evaluation of DGX-1 was performed at Fujitsu laboratories.
2. Storage system
The file system runs FUJITSU Software FEFS, high-performance scalable file system software, on six FUJITSU Server PRIMERGY RX2540 M2 PC servers, eight FUJITSU Storage ETERNUS DX200 S3 storage systems, and one FUJITSU Storage ETERNUS DX100 S3 storage system to provide the IO processing demanded by deep learning analysis.
Comments from Jim McHugh, VP and General Manager at NVIDIA
“NVIDIA DGX-1, the world’s first all-in-one AI supercomputer, is designed to meet the enormous computational needs of AI researchers. Powered by 24 DGX-1s, the RIKEN Center for Advanced Intelligence Project’s system will be the most powerful DGX-1 customer installation in the world. Its breakthrough performance will dramatically speed up deep learning research in Japan, and become a platform for solving complex problems in healthcare, manufacturing and public safety.”