Coding strategies and performance analysis of GPU accelerated image compression

Graphics Processing Units (GPUs) are freely programmable massively parallel general purpose processing units and thus offer the opportunity to off-load heavy computations from the CPU to the GPU. One application for GPU programming is image compression, where the massively parallel nature of GPUs promises high speed benefits. However, measurements with competative highly optimized CPU implementations show that GPU based codes are usually not considerably faster, or perform only with less than ideal rate-distortion performance. This article presents the predicaments of data-parallel image coding by first presenting a series of theoretical arguments that limit the performance of such implementations before advancing to existing GPU implementations demonstrating the challenges of parallel image coding. It will be argued and seen on experiments that either parts of the entropy coding and bitstream build-up must remain serial, or rate-distortion penalties must be paid when offloading all computations on the GPU.