high performance computing on graphics processing units: hgpu.org

With the introduction of many-core GPUs, there is widespread interest in using GPUs to accelerate non-graphics applications such as bioinformatics, energy, finance and several research areas. Even though the GPUs provide highly parallel processing capability, the communication interface between CPU and GPU could be a performance bottleneck due to heavy data transfer. If data transfer time is overwhelming the computation time on GPU, it would be better keep the computation on CPU instead of using GPUs. In this paper, we characterize the HMMER 3.0 and investigate performance hotspot functions. The HMMER is a bioinformatics application which is used in searching sequence databases for protein sequences. For our experiment, we use Nvidia CUDA that abstracts the GPU hardware. Based on the hotspot analysis of HMMER 3.0, we consider two factors for partial CUDA acceleration: one is the performance impact of major hotspot functions and the other one is data transfer overhead. Also, we verified that hotspot analysis based partial CUDA acceleration could provide better performance than full CUDA implementation.