Accelerator Benchmark Suite Using OpenACC Directives

In recent years, GPU computing has been very popular for scientific applications, especially after the release of programming languages like CUDA, OpenCL, and OpenACC. The growing popularity of GPU computation in commercial and scientific fields is attributed to the high computational power of GPU cores. The accelerator benchmark suite using OpenACC 2.0 is a combination of very popular benchmarks – the Parboil and NAS Parallel benchmarks. These benchmarks contain a wide range of throughput computing applications, which are useful for studying the performance of computing architectures and compilers. The Parboil benchmark includes applications from different scientific and commercial fields including image processing, biomolecular simulation, and astronomy. The NAS Parallel benchmark has a set of applications that target different areas of computational fluid dynamics. The accelerator benchmark suite that has been designed exploits the computational power of GPU architecture by using the emerging directives and clauses provided by OpenACC 2.0. This benchmark can act as a reference point for new programmers in GPU computing, reducing the time taken to understand one of the most powerful parallel programming paradigms. Finally, the goal of the accelerator benchmark is to evaluate the applicability of one of the high-level programming models OpenACC for accelerators. This benchmark will help provide the OpenACC community with valuable feedback to improve the model further.