CUDA架构下分子动力学模拟的高速实现  

High speed molecular dynamics simulation approach based on CUDA

作　　者：陈捷捷[1] 王彦浩[2] 刘丹[3] 

机构地区：[1]中国舰船研究设计中心,湖北武汉430064 [2]清华大学计算机科学与技术系,北京100083 [3]武汉第二船舶设计研究所,湖北武汉430064

出　　处：《机械》2013年第12期73-76,共4页Machinery

摘　　要：统一计算设备架构(CUDA)的提出和图形处理器(GPU)的并行处理能力和数据传输能力,使得基于CUDA架构的GPU高性能计算迅速成为一个研究热点。针对含有大规模分子动力学模拟效率低下问题,提出了一种CUDA架构下分子动力学模拟的新方法,给出了分子动力学模拟的CUDA算法流程,并利用CUDA C实现了并行算法。实验结果表明,与CPU处理速度相比,GPU可以提高速度200倍左右。The Compute Unified Device Architecture （CUDA） and the rapidly promoting ability of data-parallel processing and data transferring of Graphic Processing Units （GPU） make an attractive area for high performance computation based on CUDA. As the efficiency of large scale molecular dynamics simulation is relative low, a new approach to the molecular dynamics simulations based on CUDA is proposed. The CUDA-based algorithm of molecular dynamics simulation is give. We have used the CUDA C to implement parallel algorithm. The results indicated that the performance of the GPU is compared to their main processor counterpart. We achieve speedups of up to near 200 folds.

关 键 词：图形处理器 统一计算设备架构 分子动力学模拟 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]