基于GPU的分子动力学势函数及结构特征量模拟系统-计算机科学与技术专业论文.docxVIP

哈 哈 尔 滨 工 业 大 学 工 学 硕 士 学 位 论 文 Abstract GPU (Graphics Processing Unit) has a strong parallel computing capabilities, espe- cially CUDA (Compute Unified Device Architecture) makes the GPU appear to occupy an increasingly important position in High-Performance Computing. Molecular Dynamics simulation technique is the use of a computer system to simulate the motion of individual particles. As the simulation time is very long due to its large amount of calculation, it，s considered to compute Molecular Dynamics Simulation into GPU. Based on the analysis of parallelism of GPU and the principle ofMolecular Dynamics simulation, the most time-consuming process simulation function to calculate the potential move to GPU computing. With analyzing potential function to calculate the intrinsic par- allelism using CUDA threads and particles the one relationship, combined with the GPU with its own characteristics, to accelerate the completion of the calculation. This paper implements two pair potentials (Morse potential, Lennard-Jones potential), and two multi- body potentials (Tersoff potential, EAM potential) on GPU accelerated computing. The results show that using N V ID IA GeForce GTX 650，s GPU, relative to Intel Core2 E7500 CPU, the acceleration ratio of potential function to calculate is between 22 to 91, which also will be able to greatly shorten the time of calculation of entire molecular dynamics simulations. This paper also describes acceleration of two structural characteristics(radial distribution function(RDF), and static structure factor(SSF)), the results show that under the same platform with the above, the acceleration ratio is between 22 to 27. Finally, under the Windows platform, using Qt graphical user interface design, this paper builds a molecular dynamics simulation system, in which the amount of the poten- tial function and structural characteristics of GPU-based computing to run, and the system was verified. Keywords: GPU, CUDA, Molecular Dynamics Simu