Building a supercomputer with 1PFLOPS of peak computing performance, using many-core processors
By   |  October 20, 2015

    Challenges
  • Expanding needs of the high-speed system to support ‘Interdisciplinary Computational Science’, a fusion of science and computer science.
  • Accumulation of technology of many-core processors
    Solution
  • Intel Xeon processor E5 family
  • Intel Xeon Phi coprocessor
  • Comprehensive tool ‘Intel Cluster Studio XE’ towards MPI developers and C++/ Fortran programmers
    Benefits
  • Introduction of supercomputer ‘COMA (PACS-IX)’ of peak computing performance of 1.001PFLOPS
  • Research and develop tuning and coding of next generation supercomputers that employ many-core processors

Starting to use the second PFLOPS supercomputer at the Centre for Computational Sciences, University of Tsukuba Center for Computational Sciences, University of Tsukuba (hereinafter, Center for Computational Sciences), promoting “Interdisciplinary Computational Science”, a fusion of science and computer science, introduced “COMA (PACS-IX)” (COMA, PACS- Nine), a supercomputer of peak computing performance of 1.001PFLOPS, and has been in use since April 2014. The CPU comes with two units of Intel Xeon Processor E5 family, the arithmetic accelerator which assists the CPU, combined with 393 units of compute nodes that comes with two units of Intel Xeon Phi coprocessor, form an Ultra parallel PC cluster. It is a matter of pride to achieve the highest domestic performance (as of June 2014), with peak computing performance of 1.001PFLOPS, with supercomputers having Intel Xeon Phi coprocessors.

At the Center for Computational Sciences, in addition to promoting state-of-the-art computational science research, the two supercomputers, ‘HAPACS’, the supercomputer that came into use in 2012 with 1.116 PFLOPS that accelerates the computing device GPU, and the new ‘COMA’ are used towards the professional development of high-performance computing in the education of students and graduate students.

Self-promotion of scientific research by using a unique computer at the National University Information Centre
With the Centre for Computational Physics that was founded in 1992 as its predecessor, the current Center for Computational Sciences had its restructuring and expansion done in 2004. In 2010, it was certified by the Joint Usage/Research Center, Ministry of Education as ‘AISCI (Advanced Interdisciplinary Computational Science Collaboration Initiative)’.

Currently, under the ‘Multidisciplinary Joint-use Program’ that strengthens the collaboration of computer science and science, it provides the computer center to research institutes across the country, supports holding of workshops for interdisciplinary computational science promotion, supports invitation to researchers, supports exchange of students and researchers and conducting of research aimed at the development of extra scale supercomputers.

Deputy Director and Chairman of the Computer System operation, Mr. Taisuke Boku, explains thus: “Its role is not just that of the information center where researchers from the field of science can use supercomputers but being a unique National University with a large number of teachers and researchers from various computational science fields, application development and scientific research using this computer, and promoting the development of the computer itself, a feature that is not present in the computer centers of other universities. These initiatives are defined as ‘Interdisciplinary Computational Sciences’, and as a major feature of the Center for Computational Sciences it is a place of exploration in computational sciences that have a wide scope.” ‘Interdisciplinary Computational Science’ uses a high-speed network and ultra-highspeed computers as the main research tool, where researchers from three fields, viz. science researchers who use the computer, information science researchers who research on data and media processing, and computer science researchers who research on hardware, software, algorithms, and programming use these computers for their integrated but innovative applied research. The Computational Science Research Center, along with building a joint research system over the years, has also been promoting basic science, high-speed simulation, largescale data analysis, and applied research in information technology. Currently, using ‘HA-PACS’ and ‘COMA’, that brought into reality the PetaFLOPS, we aim for a breakthrough in the research and development of computational sciences in the fields of particle physics, space, nuclear physics, material, science, life, and global environment.

Evaluating the high cost performance and the power performance and adopting of Intel Xeon Phi coprocessor
In the recent years, speed and performance of CPU have helped performance of massively parallel PC clusters to increase significantly. However, the conventional way of just increasing the number of nodes by stacking CPU to increase computing performance has limitations due to limited space and power. At the Computational Science Research Center, a course concerning the study of low power high performance supercomputers accelerated by arithmetic devices that assist CPU focuses on many-core processors (MIC) in which multiple cores are integrated in a single chip. Intel has been participating in the ‘Intel MIC Beta Program’ initiated for evaluation of the architecture of MIC and continuously researching on tuning and performance evaluation. The research center has decided to introduce COMA instantaneously after the Intel Xeon Phi coprocessor is launched.

The Intel Xeon Phi coprocessor has 61 CPU cores on a single chip. Just attaching this coprocessor to a CPU through PCI Express (generic bus) will achieve a high cost performance as well as bring out the performance of teraflops. This coprocessor gives an excellent space performance and high power performance for every 1 watt. (Mr. Boku) Another benefit of the Intel Xeon Phi coprocessor, which shares same architecture as that of the Intel Xeon processor, is its high level of programmability. . Mr. Boku says, “There comes a sense of security when programs can be developed without wasting the existing programs and without having to remember new things. Just like the conventional coprocessors, this coprocessor can be used to write programs by using FORTRAN or C++ and Open MP can be used, so that developing programs for the future is possible” In 2013, Computational Science Research Center had designed the next supercomputer system in collaboration with the information Technology Center of the University of Tokyo and has created ‘JCAHPC (Joint Center for Advanced High Performance Computing)’ an organization for co-operation and management. As the goal for FY 2015, JCAHPC is planning to introduce very large supercomputers with scores of PFLOPS
class, at the Kashiwa campus of the Information Technology Center, University of Tokyo. COMA, which is provided with many-core processors, also plays a role as an experimental system for carrying out the research and development of tuning and coding of the next generation supercomputer.

Navigation

<12>

© HPC Today 2019 - All rights reserved.

Thank you for reading HPC Today.

Express poll

Do you use multi-screen
visualization technologies?

Industry news

Brands / Products index