A case study on porting scientific applications to GPU/CUDA