Guide to the best and worst HPCMP systems for executing individual TI-04 benchmarking applications and synthetics

Since time-to-solution and processor scalability of an application can vary greatly from one architecture to another, it is important to consider the suitability of each system with respect to that application in order to make efficient use of available resources. Given that the full set of possible applications is quite large, this paper focuses on those applications within the FY-04 technology insertion (TI-04) benchmarking suite - AERO, COBALT-60, GAMESS, HYCOM, NAMD, OOCore, and RF-CTH - for which all except AERO have a standard and large test case. Both the overall performance and performance per processor of the High Performance Computing Modernization Program´s (HPCMP´s) major systems are analyzed for each test case in order to provide general selection guidance to users and a unique perspective for code developers.