A Comparative Study of OpenMP and CUDA parallel architecture using Rodinia Benchmark Suit

From the last few decades the microprocessor industry is rapidly shifted to multi-core architectures, much of the research is going on for faster computation in conventional computer systems. Scientific application takes longer time to execute as it contains more mathematical calculation, there is need to parallelise the codes in much efficient way. Many algorithms have been proposed to parallelise the code effectively, but still there are certain limitations that prevent scalable implementations of many well known algorithms. In this research work we have presented a comparative study of CUDA, OpenMP by using two well known scientific applications of Rodinia benchmark suit, a benchmark suite for heterogeneous computing. We are comparing the execution times for serial, OpenMP and CUDA codes of LavaMD and kmeans. We found that CUDA can be utilized effectively to parallelise the application.