high performance computing on graphics processing units: hgpu.org

Programmable graphics processing unit (GPU) has over the years become an integral part of today’s computing systems. The GPU use-cases have gradually been extended from graphics towards a wide range of applications. Since the programmable GPU is now making its way to mobile devices, it is interesting to study these new use-cases also there. To test this, we created a programming environment based on the embedded profile of the fresh Khronos OpenCL standard and ran it against an image processing workload in a mobile device with CPU and GPU back-ends. The early results on performance and energy consumption with CPU + GPU configuration were promising but also suggest there is room for optimization.