skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Pentium Pro inside: I. a treecode at 430 Gigaflops on ASCI Red; II. price/performance of $50/Mflop on Loki and Hyglac

Conference ·
OSTI ID:629329
 [1];  [2]; ;  [3];  [1];  [4]
  1. Los Alamos National Lab., NM (United States)
  2. Goddard Space Flight Center, Greenbelt, MD (United States)
  3. California Inst. of Tech., Pasadena, CA (United States). Center for Advanced Computing Research
  4. Universite Catholique de Louvain, Louvain (Belgium). Mechanical Engineering Dept.

As an entry for the 1997 Gordon Bell performance prize, we present results from two methods of solving the gravitational N-body problem on the Intel Teraflops system at Sandia National Laboratory (ASCI Red). The first method, an O(N{sup 2}) algorithm, obtained 635 Gigaflops for a 1 million particle problem on 6800 Pentium Pro processors. The second solution method, a tree-code which scales as O(N log N), sustained 170 Gigaflops over a continuous 9.4 hour period on 4096 processors, integrating the motion of 322 million mutually interacting particles in a cosmology simulation, while saving over 100 Gigabytes of raw data. Additionally, the tree-code sustained 430 Gigaflops on 6800 processors for the first 5 time-steps of that simulation. This tree-code solution is approximately 105 times more efficient than the O(N{sup 2}) algorithm for this problem. As an entry for the 1997 Gordon Bell price/performance prize, we present two calculations from the disciplines of astrophysics and fluid dynamics. The simulations were performed on two 16 Pentium Pro processor Beowulf-class computers (Loki and Hyglac) constructed entirely from commodity personal computer technology, at a cost of roughly $50k each in September, 1996. The price of an equivalent system in August 1997 is less than $30. At Los Alamos, Loki performed a gravitational tree-code N-body simulation of galaxy formation using 9.75 million particles, which sustained an average of 879 Mflops over a ten day period, and produced roughly 10 Gbytes of raw data.

Research Organization:
Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Sponsoring Organization:
National Aeronautics and Space Administration, Washington, DC (United States)
DOE Contract Number:
W-7405-ENG-36
OSTI ID:
629329
Report Number(s):
LA-UR-97-3456; CONF-9706166-; ON: DE98000260; TRN: AD-a337 792
Resource Relation:
Conference: Supercomputing `97, Las Vegas, NV (United States), 30 Jun - 3 Jul 1997; Other Information: PBD: Jun 1997
Country of Publication:
United States
Language:
English