New benchmark for evaluating computer systems with a floating-point accelerator
The High-Performance Group of the Standard Performance Evaluation Corp (SPEC/HPG) is working on a new benchmark for evaluating computer systems with a floating-point accelerator.
'Systems with a floating-point accelerator, such as a general-purpose GPU or other kind of co-processor, are becoming much more prevalent in the high-performance computing arena,' says Kalyan Kumaran, SPEC/HPG chair. 'We believe a new benchmark that enables vendors, researchers and users to test and compare these systems will be welcomed by the HPC community.'
Initial candidates for the benchmark suite include Parboil from the IMPACT Research Group of the University of Illinois at Urbana-Champaign and Rodinia from the University of Virginia. The subcommittee will conduct a search programme to identify additional candidates.
'The Parboil benchmarks embody more than three years of algorithm, programming and optimisation research and education in GPU computing,' says Professor Wen-mei W. Hwu, director of the IMPACT Research Group. 'We are proud to offer these benchmarks to help jump-start the SPEC accelerator effort.
SPEC’s benchmark is expected to support multiple programming models and will come with run and reporting rules that allow fair comparisons of systems, a traditional strong point of all SPEC benchmarks.
'A standardised benchmark for evaluating computer systems with a floating-point accelerator will be a major asset to us when acquiring new systems and assessing existing ones,' says Dieter an Mey, HPC team leader for the Center for Computing and Communication at RWTH Aachen University. 'We welcome SPEC’s expertise in this area.'
Current members of the subcommittee comprise a cross-section of industry and academia, including AMD, Argonne National Laboratory, Fujitsu, IBM, Indiana University, Intel, Oracle, PGI, QLogic, SGI, Technical University of Dresden, the University of Illinois at Urbana-Champaign, and the University of Virginia.