Pushing the Limits of the Maya Supercomputer with the HPCG Benchmark

Abstract

Parallel architectures and algorithms sit at the forefront of high performance computing as ways to decrease the execution time of a computationally intense problem. Parallel architecture is hardware that has multiple cores or multiple threads, and a parallel algorithm is a software algorithm designed to be able to use multiple cores or threads simultaneously. The supercomputer maya in the UMBC High Performance Computing Facility (HPCF) is designed to provide a resource for the researchers of various disciplines who require a powerful parallel computer to solve the problems they encounter in their work. Using the newly developed High Performance Conjugate Gradient (HPCG) benchmark (www.hpcg-benchmark.org) from Sandia National Laboratories, this effort identified several runtime optimizations that allow for maximum performance on maya. These optimizations nearly doubled the reported performance of the benchmark from previous tests. A total throughput of 450 GFLOP/s was achieved using only a fourth of the hardware available on maya.