多核CPU和GPU加速分子动力学模拟  

Accelerated molecular dynamics simulation using multi-core CPU and GPU

作　　者：林江宏[1] 林锦贤[1,2] 吕暾[2,3] 

机构地区：[1]福州大学数学与计算机科学学院,福州350108 [2]福州大学福建省超级计算中心,福州350108 [3]福州大学生物科学与工程学院,福州350108

出　　处：《计算机应用》2011年第3期843-847,共5页journal of Computer Applications

基　　金：福建省高校科研专项重点项目(JK2009002);福建省科技厅青年人才基金资助项目(2008F306010107)

摘　　要：在多核中央处理器(CPU)—图形处理器(GPU)异构并行体系结构上,采用OpenMP和计算统一设备架构(CUDA)编程实现了基于AMBER力场的蛋白质分子动力学模拟程序。通过合理地将程序划分为CPU单线程、CPU多线程和GPU多线程执行部分,高效地利用了计算机的处理能力。性能测试结果表明,相对于优化后的CPU串行计算,多核CPU-GPU异构并行计算模型有强大的性能优势,特别是将占整个程序执行时间90%的作用力的计算移植到GPU上执行,获得了最高可达12倍的计算加速比。On the heterogeneous architecture of multi-core Central Processing Unit （CPU） and Graphic Processing Unit （GPU）, the Open Multi-Processing （OpenMP） and the programming interfaces of Compute Unified Device Architecture （CUDA） were used to implement a molecular dynamics simulation program based on AMBER force field. In order to efficiently use computer processing power, the program was divided into different parts which were processed by CPU single-thread, CPU multi-thread and GPU multi-thread respectively. The experimental results show that compared with the optimized CPU-based implementations, the heterogeneous parallel computing model based on multi-core CPU-GPU gets powerful performance advantage. Especially, the calculations of forces, which account for more than 90% of processing time, get at most 12 times faster than CPU-based implementations while being implemented on GPU.

关 键 词：分子动力学 图形处理器 多核中央处理器 AMBER力场 计算统一设备架构 OPENMP 

分 类 号：TP399[自动化与计算机技术—计算机应用技术] O641[自动化与计算机技术—计算机科学与技术]