N-body Astrophysical GPU Application

A GPU cluster (Dirac, NERSC) was used for benchmarking and testing of a parallel N-body Hermite Integration code for GPUs called phi-GPU.

Results show a speedup of ~ 50% by using Fermi C2050 cards (Dirac, NERSC), in comparison to Tesla C1060 cards (Laohu cluster, NAOC, Beijing). A speed up for a 6th order Hermite integration scheme, in comparison to the 4th Hermite scheme has been also achieved. Benchmarks and tests of a standard N-body astrophysical problem have been successfully published (ISC 2011 PRACE Award, Hamburg, Germany), and real astrophysical applications are being currently performed.

This work is being accomplished in the frame of the project ISAAC (Infrastructure for Astrophysics Applications Computing), and in collaboration with partners at the University of Heidelberg, Germany, and the National Astronomical Observatories of the Chinese Academy of Sciences, China, which constitute, together with the Lawrence Berkeley National Laboratory (LBNL), the International Center for Computing Science (ICCS)

1. Code Description
2. Performance
3. Scientific Applications