论文部分内容阅读
作为高性能科学计算的典型应用,利用GPU并行加速分子动力学模拟是2007年以来计算化学领域高性能计算的热点。本文概述了支持GPU加速的不同MD软件的特点和其研究进展,重点分析了Amber、GROMACS、ACEMD三个代表性软件的单GPU卡和多GPU卡计算性能,结果表明在配置相同数目GPU卡的情况下,单节点比多节点在计算性能上较有优势,桌面工作站配多块GPU卡是性价比相对较好的MD模拟计算模式。本文还考察了单精度和双精度GPU加速MD的模拟计算结果的准确性,与CPU的计算结果进行了比较,结果表明,GPU的计算结果总体而言是可信的。最后,本文对GPU并行加速MD模拟的研究现状进行总结并对未来发展做了展望。
As a typical application of high-performance scientific computing, GPU parallel accelerated molecular dynamics simulation is the hot point of high performance computing in computational chemistry since 2007. This paper outlines the features and research progress of different MD softwares that support GPU acceleration, and analyzes the computing performance of single GPU card and multi-GPU card based on Amber, GROMACS and ACEMD software. The results show that when computing the same number of GPU cards In the case, a single node is more advantageous than a multi-node in computing performance. A desktop workstation with multiple GPU cards is a relatively cost-effective MD simulation computing model. In this paper, we also examine the accuracy of the simulation results of single-precision and double-precision GPU acceleration MD and compare with the results of CPU. The results show that the calculation results of GPU are generally credible. Finally, this paper summarizes the current research status of GPU parallel accelerated MD simulation and prospects the future development.