A Performance Criteria for parallel Computation on basis of block size using CUDA Architecture