Performance Of The Suprenum-1 Supercomputer With The Full Bus Interconnection Network

INTRODUCTION We have previously described the implementation of a fluid dynamical benchmark on the SUPRENUM-1 parallel computer in 1 . The benchmark, the Shallow Water Equations, is frequently used as a model for both oceanographic and atmospheric circulation. Performance of the benchmark was measured in December 1991 and the results were reported on in detail in 1 . Those measurements utilized the original 4-ring interconnection topology. We describe here only the changes in performance which have been measured since that time and which are due primarily to the recent rewiring of the SUPRENUM busses to provide both double matrix (January 1992) and fully interconnected topologies (also called diagonal topology) (June 1992). We also remark that a small improvement in single node performance (from 4.95 Mflops to 5.33 Mflops) occurred over this time due to compiler modifications and the rewriting of one line of code to avoid a compiler error.<