New Implementations and Results for the NAS Parallel Benchmarks 2.
William C. Saphir, Rob F. Van der Wijngaart, Alex Woo, Maurice Yarrow · 1997
We present new implementations and results for the NAS Parallel Benchmarks 2 suite. The suite currently consists of seven programs. Of these LU, SP, BT, MG and FT have previously been released. Here we describe implementations of EP and IS, as well as a rewritten version of FT that corrects some problems with the original release. Performance results are shown for several architectures: IBM SP Wide Nodes / 66 MHz, IBM SP Thin Nodes 2/ 120 MHz, SGI Origin 2000, Hitachi SR 2201, Sun Ultra Enterprise 4000, Cray J90, Cray T3D, HP Exemplar-S. 1 Introduction The NAS Parallel Benchmarks (NPBs)[1, 2] are a widely-recognized suite of benchmarks originally designed to compare the performance of highly parallel computers with that of traditional supercomputers. The NPBs are specified algorithmically, and are implemented mainly by computer vendors, using techniques and optimizations appropriate for their products. Performance results based on these optimized proprietary implementations--- ...