Parallel Monte Carlo Integration Algorithm Based on GPU

摘要

Big data comes with the increasing maturity of computer network technology and its application popularity. How to process big data and mine its potential value is a hot issue today. Artificial intelligence is one of the main sources of algorithms for dealing with big data, while parallel computation is the only choice to improve computing performance. Monte Carlo method plays an important role in the application of artificial intelligence, such as computer game AlphaGo, and it requires a lot of random data to calculate, so we use it, in this paper, to explore the general method of parallel processing big data. According to Monte Carlo method, the calculation time of the algorithm will increase exponentially with the increase of the calculation accuracy. As the calculation accuracy will increase accordingly when the number of points increases. In order to improve the computing performance, a Monte Carlo parallel algorithm based on GPU-CUDA programming model is designed. The quality of the random number generator is the key factor affecting the accuracy of Monte Carlo calculation, so the higher quality Mersenne Twister random number algorithm is chosen. Taking $pi$ and e as examples, the experimental results show that the GPU-based Monte Carlo parallel algorithm is effective. This method improves both the calculation performance and the calculation accuracy.