Abstract:

The general-purpose graphic processing unit (GPGPU) has become one of the major hardware platforms for parallel computing nowadays. Open computing language (OpenCL) is an open and royalty-free standard for cross-platform parallel programming of modern processors, including graphic processing units(GPU), etc. A novel OpenCL micro benchmark suite is introduced, which is used to evaluate the software and hardware performance of GPGPU architecture. The micro benchmark suite tests the single precision floating-point capability of GPGPU, the reading and writing bandwidth and the access patterns of all memory types in the GPU memory hierarchy, etc. These OpenCL micro benchmarks have high reference values to both the OpenCL and GPGPU programmers and software architecture designers.