Performance Evaluation of Global Sequence Alignment Using CUDA Compatible Multi-Core GPU

Abstract

GPU is becoming a competitive general purpose computational hardware platform against CPU in the recent few years because of its high performance and capabilities. Recent im- provements of GPUs highly parallel programming capabilities such as CUDA has lead to a wide variety of complex application with tremendous performance improvements and this attempt on GPU leads to the next generation of high performance computing called GPGPU. Genetic Sequence alignment is considered to be one of the application domains which require further improvements in the execution speed, because it is still a computationally intensive task with increased database size. So we focus on using the massively parallel architecture of GPU as a solution for the improvement of sequence alignment task. For that purpose we have implemented a CUDA based heterogeneous parallel solution for the global sequence alignment task with Needleman-Wunsch dynamic programming algorithm. At the same time we have concentrated on di erent implementations with di erent levels of memory access patterns to identify a better parallelization strategy. Our experiment results show that the massively parallel architecture of GPU is very capable of adding better speed-up for the sequence alignment task.