Tsuyoshi Hamada, PhD

HamadaT200xAbout the speaker

Tsuyoshi Hamada is the head of the massively parallel computing division and deputy director of the Nagasaki Advanced Computing Center (NACC) and associate professor at Nagasaki University.

Tsuyoshi Hamada obtained his PhD in 2006 from the University of Tokyo. From 2006 to 2008 he was a Special Postdoctoral Researcher at RIKEN. In 2008 he became an assistant Professor at Nagasaki University, and was promoted to Associate Professor in 2010.

He has worked on massively parallel architectures based on FPGAs and GPUs. In 2004 he wrote a compiler PGR for FPGAs to automate the programming of reconfigurable systems. In 2005 he designed a FPGA-based computer PROGRAPE-4 and achieved 243 GFlops/board and 7.7 TFlops for the entire system.

Since 2006, he has been focusing on GPUs. In 2007 he released a CUDA library for N-body problems, CUNBODY. His cluster in Nagasaki started out in 2008 with 32 8800GT GPUs. In April ’08 he extended this to 128 8800GTS cards, and in November ’08 to 256 8800GTS cards. In April ’09, he acquired another 256 9800GTX cards, and in August ’09 another 288 of the GTX295 cards.

With his large GPU cluster (DEGIMA), he achieved a sustained performance of 158 TFlops and $7.2/GFlops, which led to the Gordon Bell prize for price/performance at SC’09. In 2010, he changed the interconnect to Infiniband and achieved 190 TFlops. His paper has been selected as a Gordon Bell finalist for SC’10.

His research covers:

  • special purpose computers (GRAPE)
  • FPGAs
  • GPUs
  • gravitational N-body simulations
  • fast algorithms (Treecode)

Sample publications

  • T. Hamada, R. Yokota, K. Nitadori, T. Narumi, K. Yasuoka, M. Taiji, and K. Oguri, 42 TFlops Hierarchical N -body simulations on GPUs with Applications in both Astrophysics and Turbulence. Proceedings of the Conference on High Performance Computing Networking, Storage and Analysis, pp. 1-12, (2009)
  • T. Hamada, K. Nitadori, K. Benkrid, Y. Ohno, G. Morimoto, T. Masada, Y. Shibata, K. Oguri, and M. Taiji, A Novel Multiple-walk Parallel Algorithm for the Barnes-Hut Treecode on GPUs- Towards Cost Effective, High Performance N -body Simulation. Computer Science – Research and Development, Special Issue Paper, pp. 1-11, (2009)