earticle

영어

Ultra high definition (UHD) game scenes have caused the memory bandwidth problem. The lossless DPCM-GR based compression algorithm [12] using NVIDIA CUDA(Compute Unified Device Architecture) like general purpose GPU (GPGPU) computing relieves the bandwidth problem without sacrificing image quality, which supports bit parallel pipelining. This paper increases the memory bandwidth efficiency using the shared memory of CUDA based on the compression algorithm [12]. Also, various asynchronous transfer configurations which can overlap the kernel execution and data transfer between the Host and the CUDA device are implemented with the page-locked host memory. Experimental results show that GPGPU CUDA computing obtains the maximum 87.5 and 30.6 times speedups for GTX650Ti and GT330, respectively, comparing to Host CPU. Also, the maximum reductions of the compression time for GTX650Ti and GT330 are 54.1% and 30.3%, respectively, among various concurrency transfer configurations.