计算机工程

• 开发研究与工程应用 • 上一篇    下一篇

基于GPU 的碳纳米管分子动力学并行仿真

孟小华a,覃大胜a,郑冬琴b ,周玉宇a   

  1. (暨南大学a. 计算机科学系;b. 物理系,广州510632)
  • 收稿日期:2014-04-09 出版日期:2015-04-15 发布日期:2015-04-15
  • 作者简介:孟小华(1965 - ),男,副教授、硕士、CCF 会员,主研方向:并行分布式系统;覃大胜,硕士研究生;郑冬琴,副教授、博士; 周玉宇,副研究员,博士。
  • 基金项目:
    国家“863”计划基金资助项目(2013AA040404);广东省重点实验室建设基金资助项目(2011A060901026)。

Parallel Simulation of Carbon Nano Tube Molecular Dynamics Based on GPU

MENG Xiaohua a ,QIN Dasheng a ,ZHENG Dongqin b ,ZHOU Yuyu a   

  1. (a. Department of Compute Science;b. Department of Physics,Jinan University,Guangzhou 510632,China)
  • Received:2014-04-09 Online:2015-04-15 Published:2015-04-15

摘要: 基于计算机的分子动力学仿真具有理论分析方法和实验方法无法比拟的优点,但分子动力学仿真算法计算量非常大,特别是在对碳纳米管的大规模粒子数进行仿真处理时,普通的基于CPU 的串行算法执行效率低且耗时多。为此,提出基于统一计算设备架构的碳纳米管分子动力学的图形处理单元(GPU)并行算法,设计并实现仿真算法中适合GPU 并行运算的分裂算法,将具有竞争资源的运算以非竞争方式运行。实验结果表明,与CPU 串行 仿真算法相比,分裂算法的运算速度较快,且在只有16 个GPU 流处理器显卡上可获得十多倍的加速比。

关键词: 碳纳米管, 分子动力学, 统一计算设备架构, 并行计算, 时间效率, 加速比

Abstract: Molecular dynamics simulation has advantages dramatically superior to both theoretical analysis and experiments. However,due to the extremely high cost of computation resources during the simulation of a large number of Carbon Nano Tube(CNT) particles,typical CPU serial algorithm implementation is non-efficient and slow. A Compute Unified Device Architecture(CUDA) based parallel algorithm of CNT molecular dynamics is proposed in this paper to take advantage of the data parallelism of Graphic Processing Unit (GPU). A CNT is divided to several blocks and processed parallel in the GPU. Experimental results show that the algorithm can obtain a speed-up more than 10 times to the CPU serial algorithm in a low-configured graphics card that has only 16 GPU stream processors.

Key words: Carbon Nano Tube(CNT), molecular dynamic, Compute Unified Device Architecture(CUDA), parallel computation, time efficiency, speed-up ratio

中图分类号: