TOKYO, June 20, 2017 – Fujitsu Limited and Fujitsu Taiwan Limited today announced that they will build a new supercomputer system for Taiwan’s National Center for High-performance Computing, National Applied Research Laboratories (NCHC) that will have the highest performance in the country at the time it commences operations, scheduled for May 2018. The new system will serve as the core platform for research and development in Taiwan, fostering the development and growth of Taiwan’s overall industries and economy.
The new PC cluster system is comprised of 715 servers, including a next-generation model of the Fujitsu Server PRIMERGY, and brings together the most advanced high-performance computing technologies, such as the use of the latest server CPU from Intel and fastest current GPU from NVIDIA. The CPUs have a theoretical peak performance of over 2.13 petaflops(1), and the GPUs have a theoretical peak performance of over 1.35 petaflops, making the expected overall theoretical peak performance of the system over 3.48 petaflops.
Background to the Deployment of the New System
Established in 1991, NCHC is Taiwan’s only national-level supercomputer center.
From Taiwan’s research institutions and private companies, as well as foreign researchers, over 1,000 people use the center, promoting international people-to-people exchanges and joint research. It thereby contributes to innovation and the advancement of Taiwan’s scientific and technological capabilities, and in turn, to the development of Taiwan’s economy.
The new system will give the center a research platform that can accommodate such new research fields as AI and big data, as well as accelerate the pace of further research.
Overview of the New Supercomputer System
The new PC cluster system is comprised of 715 servers, including a next generation model of Fujitsu’s PRIMERGY. Within this, 631 computation nodes use the Intel Xeon processor Scalable family (Codename: Skylake), and deliver a theoretical peak performance of over 2.13 petaflops. In addition, 64 of its accelerator nodes use the NVIDIA Tesla P100, a GPU from NVIDIA, for a theoretical peak performance of over 1.35 petaflops. Each computation node is connected by Fujitsu’s latest high-speed interconnect(2), Intel Omni-Path Architecture, and with the acceleration technology held by Fujitsu Limited and Fujitsu Laboratories Ltd., delivers highly parallel processing performance.
Moreover, the system uses the Fujitsu’s cutting-edge cooling technology(3), enabling energy conservation with an efficient cooling system.
(1) Petaflops
Short for peta floating point operations per second. Peta is an SI prefix indicating one quadrillion, or 10 to the power of 15, so this indicates performance of one quadrillion floating point operations per second.
(2) Interconnect
The high-speed network connecting computation nodes.
(3) Cooling technology
This technique allows for cooling with higher-temperature water than that of a conventional water cooling system, thus reducing the load on the hardware that provides cooled water. Furthermore, the heat radiated from the heated cooling water can be reused to further improve the system’s energy efficiency.
