分子动力学模拟计算在通用图形处理芯片上的实现

将在计算生物分子中广泛应用的CHARMM力场应用于Windows computer cluster server(WCCS)环境下, 并实现了该力场及分子动力学模拟程序的通用显卡(GPU)并行计算. 对一些多肽链的动力学模拟结果显示, 与CPU计算相比, GPU计算在计算速度上有巨大的提升. 与64位Athlon 2.0G相比, 在NVIDIA GeForce 8800 GT显卡上的动力学模拟速度提高了至少10倍, 而且这个效率比会随着模拟体系及每块尺寸的增大而增大. 模拟体系的增大使得GPU并行单元的计算空载相对减少, 块尺寸的增大使缓存区尺寸相对减少, 单块计算效率得以提高. 在测试样本中, 该效率比最高可达到28倍以上. 利用GPU计算还对一条含有397个原子的多肽链进行了分子动力学模拟, 给出了氢键分布随时间的变化结果.

Molecular Dynamics Simulation Using Graphics Processing Units

In this paper,the molecular dynamics program with CHARMM force field is developed on Graphics processing unit(GPU)at Windows computer cluster server(WCCS)system.From the testing results of peptide chain,the efficiency of GPU is outstanding compared with that of CPU.The efficiency on NVIDIA GeForce 8800 GT GPU is at lease 10 times faster than that on a single Athlon 2.0G CPU.When the total molecule size is enlarged,the number of vacant parallel units in GPU decreases,so the parallel efficiency increases.At the same time,while the fragment size is enlarged,the buffer size of a fragment decreases relatively,so the total efficiency also increases.The maximum efficiency ratio of GPU/CPU reaches to 28 times according to our test.At last,a peptide chain with 397 atoms is tested for simulation and the population of hydrogen bond is described at different time steps.