Abstract
Simulation of lattice quantum chromodynamics (QCD) on the 128-node Caltech/JPL lellarklllfp Hypercube achieves a sustained speed of 600 MFLOPS. The simu lations have produced excellent physics results on the forces between quarks. A speedup of 100 on the 128- node machine is obtained. We stress the importance of keeping up the communication speed as the processor speed increases and have devised a vectorized scheme to reduce the communication latency, leading to a 71% reduction in communication time. Practical experiences with the parallel implementation are discussed and a general performance analysis is reported.

This publication has 8 references indexed in Scilit: