Achievement Award 2013

Achievement Award

Research and Development of the K computer

Fumiyoshi Shoji , Yoshihiro Kusano , Mitsuo Yokokawa

Fumiyoshi Shoji

Yoshihiro Kusano

Mitsuo Yokokawa

　The K computer is a distributed-memory parallel supercomputer developed by RIKEN and Fujitsu Ltd. It brings together leading-edge technologies, including an ultra-fast CPU architecture called HPC-ACE, a direct interconnect network Tofu [1], energy efficient implementation using a novel water-cooling system, etc. As shown in Figure 1, it is a large-scale computer system comprising 88,128 CPUs, or 705,024 cores. Research and development were initiated in 2006 and, in October 2011, LINPACK performance of 10.51 petaflops was achieved; hence, K became the fastest supercomputer on the TOP500 list [3, 6, 8]. K also received top ranking in four important performance benchmarks, HPL, Global Random Access, EP Stream, and Global FFT, at the 2011 HPC Challenge Awards. The whole system was completed in June 2012 and was released, on schedule, to general users in September of that year. Since then, a variety of remarkable research results have been obtained [4, 7].
　The hardware system uses newly developed scalar CPUs (SPARC64TMVIIIfx, 8 cores, 128 gigaflops) manufactured with 45 nm CMOS process technology. The core supports 2-way SIMD (Single Instruction Multiple Data) execution of floating-point instructions. With the SIMD execution, up to 8 floating-point operations can be executed per cycle per core [2]. The processor chip also supports memory bandwidth as high as 64 GB/s. These features realize high performance on various HPC workloads. The interconnect network Tofu has also been newly developed to ensure high flexibility and scalability. The physical topology is 6-dimensional mesh/torus, and each node is connected by a 5 GB/s bidirectional link per logical direction. The network supports detouring of faulty nodes to improve usability, fault-tolerance and interoperability.
　The K computer supports a Linux-based OS. The system software adopts a two-level local/global file system based on the Lustre file system to provide exa-byte-scale storage as well as realizing efficient load balancing and job scheduling. To apply 10-petaflops high performance to a wide range of users, the system provides optimized Fortran and C/C++ language environments, and the standard MPI library [5, 9].
　As stated above, a number of leading-edge technologies, not only in hardware but also in software, were introduced in order to develop the K computer system. The product stands as powerful infrastructure that makes a valuable contribution to the ability to obtain useful results in various areas of computational science. These quite outstanding achievements qualify the recipients for recognition with the Achievement Award.
　Although a number of individuals took part in the research project, we are limited by the award bylaws to recommending only three persons as award winners.

Fig 1．The K computer

Fig 2．K computer cabinet

References

(１)　Y. Ajima, S. Sumimoto, T. Shimizu, "Tofu: a 6D mesh/torus interconnect for exascale computers," Computer, vol.42, no.11, pp.36-40 (2009)
(２)　T. Maruyama, T. Yoshida, R. Kan, I. Yamazaki, S. Yamamura, N. Takahashi, M. Hondou, and H. Okano, "Sparc64 VIIIfx: A new-generation octocore processor for petascale computing," IEEE Micro, vol.30, no.2, pp.30-40, March-April (2010)
(３)　M. Yokokawa, F. Shoji, A. Uno, M. Kurokawa, T. Watanabe, The K computer: Japanese next-generation supercomputer development project, Proc. of 17 IEEE/ACM Int. Symposium on Low-Power Electronics and Design, 371-372 (2011)
(４)　Y. Hasegawa, J. Iwata, M. Tsuji, D. Takahashi, A. Oshiyama, K. Minami, T. Boku, F. Shoji, A. Uno, M. Kurokawa, H. Inoue, I. Miyoshi, M. Yokokawa, First-principles calculations of electron states of a silicon nanowire with 100,000 atoms on the K computer, Proc. of 2011 Int. Conf. for High Performance Computing, Networking, Storage and Analysis (SC' 11), November, Seattle, 1-11 (2011)
(５)　K. Matsumoto, T. Adachi, S. Sumimoto, T. Nanri, T. Soga, A. Uno, M. Kurokawa, F. Shoji, M. Yokokawa, Implementation and Evaluation of MPI_Allreduce on the K Computer, IPSJ Trans. on ACS, 5(5), 152-162 (2012)
(６) Feature article : The supercomputer K (in Japanese), IPSJ magazine, 53(8), 752-807 (2012)
(７) Y. Hasegawa, J. Iwata, M. Tsuji, D. Takahashi, A. Oshiyama, K. Minami, T. Boku, H. Inoue, Y. Kitazawa, I. Miyoshi, and M. Yokokawa, Performance evaluation of ultra-large-scale first-principles electronic structure calculation code on the K computer, The International Journal of High Performance Computing Applications, International Journal of High Performance Computing Applications, published on October 17, 2013 as doi:10.1177/1094342013508163
(８) T. Shimizu, et al., Design and Evaluation of K Computer (in Japanese), Trans. on IEICE, J96-D (10), 2118-2129 (2013)
(９) T. Adachi, N. Shida, K. Miura, S. Sumimoto, A. Uno, M. Kurokawa, F. Shoji, and M. Yokokawa, Design of ultra scalable MPI collective communication on the K computer, Computer Science - Research and Development, 28, Issue 2-3, 147-155 (2013)