点击次数：

论文类型： 期刊论文

发表时间： 2013-11-15

发表刊物： Journal of Computational Information Systems

收录刊物： EI、Scopus

卷号： 9

期号： 22

页面范围： 9003-9011

ISSN号： 15539105

摘要： The power of high performance computing (HPC) heavily depends on the ability to efficiently enhancing huge amounts of parallelism. Random numbers or pseudo random numbers are very important for the efficient implementation for stochastic algorithms. Multi-core CPU and many-core Graphic Processing Units (GPUs) are conductive accelerator to produce the countless random numbers. Nevertheless, GPU does not support to directly call the library offered by CPU. In this paper, we present a novel but simple algorithm for high performance random number generation (called CUDA-RNG). Our experimental results show that this novel generator of RGN can achieve up to 189.32    speedup over the sequential implementation with a small memory load overhead when using 256 threads per block. ? 2013 Binary Information Press.