Pentium Pro Inside: I. A Treecode at 430 Gigaflops on ASCI Red, II. Price/Performance of $50/Mflop on Loki and Hyglac
SC Conference (1997)
San Jose, California
Nov. 15, 1997 to Nov. 21, 1997
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/SC.1997.10057
Michael S. Warren , Los Alamos National Laboratory
John K. Salmon , California Institute of Technology
Donald J. Becker , Goddard Space Flight Center
M. Patrick Goda , Los Alamos National Laboratory
Thomas Sterling , Center for Advanced Computing Research
Grégoire S. Winckelmans , Universite Catholique de Louvain
We present results from two methods of solving the gravitational N-body problem on ASCI Red. The first method, a trivial O(N^2) algorithm, obtained 635 Gflops for a 1 million particle problem on 6800 Pentium Pro processors. The second method, a treecode which scales as O(N log N), sustained 170 Gflops over a continuous 9.4 hour period on 4096 processors and 430 Gflops on 6800 processors during the initial part of the simulation. We also present two simulations which sustained roughly one Gigaflop on each of two 16 processor Beowulf-class computers constructed entirely from commodity personal computer technology for $50k each in September, 1996.
M. S. Warren, M. P. Goda, D. J. Becker, J. K. Salmon, G. S. Winckelmans and T. Sterling, "Pentium Pro Inside: I. A Treecode at 430 Gigaflops on ASCI Red, II. Price/Performance of $50/Mflop on Loki and Hyglac," SC Conference(SC), San Jose, California, 1997, pp. 61.