Concurrent Computing

Gamera Cluster Performance

Superlinear Speedup

For the table below, the time to calculate the SF6 ground state energy and two points of the polarization potential for the e+-SF6 system using the serial and concurrent versions of the PATMOL code are compared.

# ProcessorsExecution Time (minutes)
 Gamera
1222 (Serial)
1610 (MPI)

We believe that the "superlinear" speedup is due to having 16 cache lines and 16 local disks available for the two-electron integrals.


This page designed and maintained by T.L. Gibson
thomas.gibson@ttu.edu

Page Last Modified July 11, 2000

Return to Gibson Home Page