基于GPU的分子动力学势函数及结构特征量模拟系统

发布时间：2018-11-18 11:47

【摘要】：GPU(Graphics Processing Unit,图形处理器)具有很强的并行计算能力,尤其CUDA(Compute Unified Device Architecture,统一计算设备架构)的出现使得GPU在高性能计算中占据越来越重要的地位。分子动力学模拟技术是指利用计算机来模拟体系中各个粒子随时间的运动变化情况,由于其计算量较大,因此考虑放到GPU上来执行。本文在分析了GPU的并行能力及分子动力学模拟原理的基础上,将模拟计算过程中耗时最多的势函数计算移至GPU计算,分析势函数计算的内在并行性,利用CUDA线程与粒子间的一一对应关系,并结合GPU自身特点,完成计算的加速。本文实现了两种对势(Morse势、Lennard-Jones势)和两种多体势(Tersoff势、EAM势)的GPU加速计算。计算结果表明,在NVIDIA GeForce GTX650的GPU上,相对于Intel Core2E7500的CPU,势函数计算的加速比在22-91之间,这也就能极大的缩短整个分子动力学模拟计算的时间。另外本文还介绍了两个结构特征量(径向分布函数(RDF)、静态结构因子(SSF))的GPU加速,计算结果表明,在与上述相同平台下,加速比达到22-27。最后,本文在Windows平台下,采用Qt设计图形用户界面,完成了分子动力学模拟系统的搭建,其中的势函数及结构特征量的计算基于GPU运行,并对系统进行了测试。
[Abstract]:GPU (Graphics Processing Unit, (graphics processor) has a strong parallel computing capability, especially the appearance of CUDA (Compute Unified Device Architecture, unified computing device architecture, which makes GPU play an increasingly important role in high performance computing. Molecular dynamics simulation technology refers to the use of computer to simulate the movement of each particle in the system with time. Due to the large amount of calculation, it is considered to be carried out on GPU. Based on the analysis of the parallel capability of GPU and the principle of molecular dynamics simulation, the potential function calculation, which takes the most time in the simulation calculation, is moved to the GPU calculation, and the inherent parallelism of the potential function calculation is analyzed. Using the one-to-one correspondence between CUDA threads and particles and combining with the characteristics of GPU, the computation is accelerated. In this paper, the GPU acceleration calculations of two pairs of potential (Morse potential, Lennard-Jones potential) and two kinds of multibody potential (Tersoff potential, EAM potential) are realized. The results show that the acceleration ratio calculated by the CPU, potential function relative to that of the Intel Core2E7500 on the GPU of NVIDIA GeForce GTX650 is between 22-91, which can greatly shorten the time of the whole molecular dynamics simulation. In addition, the GPU acceleration of two structural characteristic variables (radial distribution function (RDF), static structure factor (SSF) is also introduced. The calculation results show that the speedup ratio is 22-27 under the same platform. Finally, in this paper, the graphical user interface is designed by using Qt on Windows platform, and the molecular dynamics simulation system is built. The calculation of potential function and structural characteristic is based on GPU, and the system is tested.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP332;TP391.41

【参考文献】