Summary of Research Project Results under JSPS FY2001
"Research for the Future Program"



1.Research Institution University of Tsukuba
 
2.Research Area Physical and Engineering Sciences
 
3.Research Field Computational Science and Engineering
 
4.Term of Project FY 1997 〜 FY 2001
 
5.Project Number 97P01102
 
6.Title of Project Development of Next-generation Massively Parallel Computers

7.Project Leader
Name Institution,Department Title of Position
Yoichi Iwasaki University of Tsukuba Vice President

8.Core Members

Names Institution,Department Title of Position
Akira Ukawa University of Tsukuba, Institute of Physics professor
Junichiro Makino University of Tokyo, Graduate School of Science associate professor
Taisuke Boku University of Tsukuba, Institute of Electronics and Information Science associate professor

9.Summary of Research Results

Rapid progress of computational sciences and engineering is deeply connected with development of massively parallel computers (MPP). This project aimed to carry out R&D to enhance the speed of MPP from 1TFLOPS to 100TFLOPS. The research was carried out for MPP for continuous physical systems and for particle-based systems, whose results were integrated toward the end of the project.
For MPP for continuous physical systems, a novel processor architecture called SCIMA was developed. Using the software controllable feature of on-chip memory, this processor is shown to realize superior performance over cache-based processors for HPC applications. A high-speed flexible parallel I/O and visualization system named PAVEMENT was also developed and installed in a wide variety of platforms from MPP to PC clusters.
For MPP for particle-based simulations, a special-purpose processor GRAPE-6 for gravity/Coulomb interactions were developed. In June 2001, the GRAPE-6 system with a peak speed of 32TFLOPS achieved 11.55TFLOPS in an astrophysical simulation for galactic nuclei, for which the Gordon Bell Prize for 2001 was awarded. The GRAPE system also won the Gordon Bell Prize of 2000 for performance of the prototype GRAPE-6, and of 1999 for cost/performance of GRAPE-5.
Simulations of complex systems having both continuous and particle degrees of freedom and multiple scales of interactions are becoming more and more important in science and engineering. As a final goal of the project, a heterogeneous multi-computer system (HMCS) consisting of the CP-PACS (0.6TFLOPS) for continuous systems and GRAPE-6(8 boards and 8TFLOPS) for particle systems connected via the parallel PAVEMENT I/O system was built. With this powerful HMCS having an integrated peak speed of 8.6TFLOPS, a galaxy formation simulation treating matter, radiation and gravity rigorously was carried out for the first time. These achievements demonstrated that HMCS provides an ideal platform for next-generation HPC in science and engineering.

10.Key Words

(1)computational science、(2)large-scale simulation、(3)massively parallel computer
(4)processor in memory、(5)parallel I/O system、(6)visualization system
(7)special-purpose processor、(8)FPGA、(9)heterogeneous multi-computer


back