Scalable Matrix Integration

Gamera Cluster Performance vs SGI O2K Cluster

For the table below, the MPI version of the 3D Direct Quadratrue DPM code was used to compute one set of modified Nuclear Attraction Integrals for one point of the positron-N2 system in a 93-primitives GTO basis with 65,536 quadrature points.

The times listed for the SGI Origin 2000 are best (not typical) times, since other users and their jobs were present during these tests. Also, it is not clear that more than 8 processors on the SGI were actually assigned for the 16-processor test case.

# ProcessorsExecution Time (seconds)
 GameraSGI Origin 2000

The parallel efficiency (1 vs 8) for the SGI Origin 2000 is 84% for this code, while for Gamera, the (1 vs 8) parallel efficiency is 95% and the (1 vs 16) parallel efficiency is 88%. With 16 nodes, the Gamera cluster achieves 1.1 Gflops on this test case.

This page designed and maintained by T.L. Gibson

Gibson Home Page Return to Gibson Home Page