SPECaccel® 2023 benchmark
The SPECaccel® 2023 benchmark benchmark suite performance with computationally intensive parallel applications running under the OpenACC and OpenMP target offloading APIs. The suite exercises the performance of the accelerator, host CPU, memory transfer between host and accelerator, support libraries and drivers, and compilers.
Benchmark tests
The benchmark suite uses many of the original benchmarks from SPEC ACCEL V1.4 with increased workload sizes and runtime to better measure modern accelerators. It contains 12 application benchmarks ported to use either OpenACC, OpenMP using "loop" constructs, OpenMP with "distribute" constructs favoring collapsing all loop levels, and OpenMP with "distribute" applied to outer loops and "simd" applied to inner loops. The choice of model is selected by the user based on the compiler and accelerator they are using. Models may be mixed in peak results.
SPECaccel 2023 includes benchmarks from NAS Parallel Benchmarks (NPB), SPEC OMP® 2012, Parboil benchmark from the IMPACT Research Group of the University of Illinois at Urbana-Champaign and others derived from high-performance computing (HPC) applications.
"Engaging students with a benchmark suite developed collaboratively by academic researchers, industry and lab partners is key as it exposes them to the orchestration of the suite, usage of directives, the intense experimentation and evaluation of results on more than just one platform, the numerous decisions that are made along the way before an application could be bundled into the suite"
—Dr. Sunita Chandrasekaran, Associate Professor, Dept. of CIS, University of Delaware
Results
Results for all of the SPECaccel® 2023 benchmark metrics; includes all of the results submitted to SPEC from the SPEC member companies and other licensees of the benchmark.