ACCMS to inaugurate a new supercomputer cluster in October (24 May 2016)

发表日期

On 24 May, the Academic Center for Computing and Media Studies (ACCMS) shared details of the supercomputer system it plans to put into service in October 2016. The cluster will be comprised of Cray Inc's XC40 and CS400 supercomputers and include DataDirect Networks's (DDN's) SFA14K storage platform.

The XC40 supercomputer consists of 1,800 compute nodes, each equipped with an Intel Xeon Phi many-core processor (codenamed Knights Landing). These nodes are interconnected via Cray's Dragonfly network topology to deliver an aggregate peak performance of 5.48PFLOPS, which will, at the time of the computer's cutover in October, be the second highest in Japan after the K-computer at the RIKEN Advanced Institute for Computational Science (AICS).

The CS400 is a high-performance cluster supercomputer consisting of 850 dual-socket compute nodes, each equipped with an Intel® Xeon® Processor E5-2695 v4 (codenamed Broadwell), interconnected by Intel's Omni-Path Architecture to offer an aggregate peak performance of 1.03PFLOPS.

These machines will both be connected to the DDN EXAScaler, a 24-PByte parallel-file storage appliance, and each integrated with the Cray DataWarp 230-TByte I/O accelerator and the DDN IME (Infinite Memory Engine) flash-caching software to enable fast read/write of large files.

ACCMS being a Joint Usage/Research Center, the new supercomputer system will be made available to all domestic universities and research institutes. It will also be offered for use for large-scale high-performance computing projects pursued in the frameworks of the Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures (JHPCN), of which ACCMS is a member, and High Performance Computing Infrastructure (HPCI).

Glossary of terms

  • Multicore processor
    A processor with up to 20 general-purpose CPU cores. Intel® Xeon® Processor E5-2695 v4, the key component of the CS400's compute node, has 18 CPU cores and offers a peak performance of more than 0.6TFLOPS.
  • Many-core processor
    A processor with several dozen CPU cores specialized for high-performance scientific computing. Intel® Xeon Phi Processor has up to 72 CPU cores and offers a peak performance of more than 3TFLOPS.
  • Peak performance, PFLOPS, TFLOPS
    "FLOPS" in PFLOPS and TFLOPS is an acronym for FLoating-point Operations Per Second, which refers to the maximum number of floating-point operations that can theoretically be performed by a processor or supercomputer in one second, used as a measure of such devices' peak performance. FLOPS can be calculated by multiplying the number of floating-point units in a device by the operating frequency of each of those units. PFLOPS and TFLOPS refer to, respectively, quadrillion (10 15 ) and trillion (10 12 ) operations per second.
  • Interconnect
    A network built for high-speed high-volume data transmission among compute nodes, examples of which include processors and memories. An interconnect network's per-node communication performance equals or exceeds that of a nationwide network backbone, while its system-wide communication performance is several hundred times higher.
  • Parallel file system
    A storage system comprised of thousands of hard disks and many controllers, used for storing a great number of large files and enabling a supercomputer to perform high-speed read and write operations on those files in parallel.
  • Ultra-high-speed storage system
    A storage system comprised of a large number of solid-state drives (SSDs) made from semiconductor materials to enable faster file read/write operations than can be done with a hard-disk-based system.
  • Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures (JHPCN)
    A network of supercomputer centers hosted by eight universities -- Hokkaido University, Tohoku University, the University of Tokyo, Tokyo Institute of Technology, Nagoya University, Kyoto University, Osaka University, and Kyushu University. Designated by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) as a "network-type Joint Usage/Research Center", JHPCN has hosted a large number of collaborative projects by researchers from member institutions and others that involved the use of the supercomputer network.
  • High Performance Computing Infrastructure (HPCI)
    A platform for high-performance computing research supported by a supercomputer network comprised of RIKEN's K-computer and others hosted by nine universities and two research institutes. A great number of strategic and collaborative research projects have been conducted using HPCI resources.

Inquiries

Academic Center for Computing and Media Studies (ACCMS), Kyoto University
Address: Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501 Japan
Phone: 075-753-7400 / Fax: 075-753-7450
Email: soumu*media.kyoto-u.ac.jp (Please replace * with @)

Related links

Latest news