Performance evaluation and tuning of GRAPE-6 - towards 40 "real" Tflops

ABSTRACT

In this paper, we describe the performance characteristics of GRAPE-6, the sixth-generation special-purpose computer for gravitational many-body problems. GRAPE-6 consists of 2048 custom pipeline chips, each of which integrates six pipeline processors specialized for the calculation of gravitational interaction between particles. The GRAPE hardware performs the evaluation of the interaction. The frontend processors perform all other operations, such as the time integration of the orbits of particles, I/O, on-the-fly analysis etc. The theoretical peak speed of GRAPE-6 is 63.4 Tflops. We present the result of benchmark runs, and discuss the performance characteristics. We also present the measured performance for a few real scientific applications. The best performance so far achieved with real applications is 35.3 Tflops.

REFERENCES

• 1: [1] J. Aarseth, Sverre. Dynamical evolution of clusters of galaxies, i. Monthly Notices of Royal Astronomical Society, 126:223-255, 1963.
• 2: [2] S. J. Aarseth. Star Cluster Simulations: the State of the Art. Celestial Mechanics and Dynamical Astronomy, 73:127-137, 1999.
• 3: [3] J. Barnes and P. Hut. A hiearchical o(nlogn) force calculation algorithm. Nature, 324:446-449, 1986.
• 4: [4] P. Chatterjee, L. Hernquist, and A. Loeb. Effects of wandering on the coalescence of black hole binaries in galactic centers, astro-ph/0302573, 2003.
• 5: Ernst Nils Dorband , Marc Hemsendorf , David Merritt, Systolic and hyper-systolic algorithms for the gravitational N-body problem, with an application to Brownian motion, Journal of Computational Physics, v.185 n.2, p.484-511, March 2003 [doi>10.1016/S0021-9991(02)00067-0]
• 6: L. Greengard , V. Rokhlin, A fast algorithm for particle simulations, Journal of Computational Physics, v.73 n.2, p.325-348, December 1, 1987 [doi>10.1016/0021-9991(87)90140-9]
• 7: [7] D. C. Heggie and R. D. Mathieu. Standardised units and time scales. In P. Hut and S. McMillan, editors, The Use of Supercomputers in Stellar Dynamics, pages 233-236, New York, 1986. Springer.
• 8: [8] T. Ito, J. Makino, T. Ebisuzaki, and D. Sugimoto. A special-purpose n-body machine grape-1. Computer Physics Communications, 60:187-194, 1990.
• 9: [9] J. Makino. An efficient parallel algorithm for O(N²) direct summation method and its variations on distributed-memory parallel machines. New Astronomy, 7:373-384, Oct. 2002.
• 10: [10] J. Makino and S. J. Aarseth. On a hermite integrator with ahmad-cohen scheme for gravitational many-body problems. Publications of the Astronomical Society of Japan, 44:141-151, 1992.
• 11: [11] J. Makino and P. Hut. Performance analysis of direct n-body calculations. The Astrophysical Journal Supplement Series, 68:833-856, 1988.
• 12: Junichiro Makino , Eiichiro Kokubo , Toshiyuki Fukushige , Hiroshi Daisaka, A 29.5 Tflops simulation of planetesimals in Uranus-Neptune region on GRAPE-6, Proceedings of the 2002 ACM/IEEE conference on Supercomputing, p.1-14, November 16, 2002, Baltimore, Maryland
• 13: Junichiro Makino , Makoto Taiji, Astrophysical N-body simulations on GRAPE-4 special-purpose computer, Proceedings of the 1995 ACM/IEEE conference on Supercomputing (CDROM), p.63-es, December 04-08, 1995, San Diego, California, United States [doi>10.1145/224170.224400]
• 14: J. Makimo , Makoto Taiji, Scientific Simulations with Special Purpose Computers: The Grade Systems, John Wiley & Sons, Inc., New York, NY, 1998
• 15: [15] J. Makino, M. Taiji, T. Ebisuzaki, and D. Sugimoto. Grape-4: A massively parallel special-purpose computer for collisional n-body simulations. The Astrophysical Journal, 480:432-446, 1997.
• 16: [16] S. L. W. McMillan and S. J. Aarseth. An o(n log n) integration scheme for collisional stellar systems. The Astrophysical Journal, 414:200-212, 1993.
• 17: [17] M. Milosavljevi¿ and D. Merritt. Formation of Galactic Nuclei. The Astrophysical Journal, 563:34-62, Dec. 2001.
• 18: [18] V. Springel, N. Yoshida, and S. D. White. Gadget: A code for collisionless and gasdynamical cosmological simulations. New Astronomy, 6:79-117, 2001.
• 19: [19] R. Spurzem and H. Baumgardt. A parallel implementation of an aarseth n-body integrator on general and special purpose supercomputers. submitted to Monthly Notices of Royal Astronomical Society, 1999.
• 20: [20] M. S. Warren, J. K. Salmon, D. J. Becker, M. P. Goda, T. Sterling, and G. S. Winckelmans. Pentium pro inside: I. a treecode at 430 gigaflops on asci red, ii. price/performance of $50/mflop on loki and hyglac. In Proceedings of SC97, pages (CD-ROM). ACM, 1997.