摘要
选择Valinomycin,Barnase等8个化学、生命科学领域的典型大分子体系,基函数数目从1160至12716范围,用NVIDIA K80 GPU高性能工作站,测试和分析了基于GPU特性设计的TeraChem量子化学程序的Hartree-Fock ab initio计算性能、并行效率,并与GAMESS程序在CPU典型集群上的计算精度和效率进行对比。结果表明,TeraChem在单一K80 GPU工作站的计算速度是16节点四核CPU集群的2至17倍;当GPU计算核心数增加4倍,GPU平台计算性能还能有2.5倍增长;若正确控制计算过程参数,GPU与CPU平台计算结果的计算精度完全一致。TeraChem在K80 GPU上的卓越计算性能和表现,可让化学家在普通化学实验室而不是超级计算中心,完成大规模的非经验量子化学计算,挑战计算化学极限。
This paper selected 8 representative molecules, including Valinomycin and Barnase, from the fields of chemistry and life science, and tested the Hartree-Fock ab initio calculation performance and efficiency of the TeraChem program on K80 GPU workstation using the number of basis sets from 1160-12716 as the benchmark. By comparing the performance of TeraChem and that of GAMESS, the results showed that the computing speed of TeraCbem based on a K80 GPU is 2 to 17 times faster than that of GAMESS based on a 16-node CPU PC-Cluster. TeraChem computing performance can be increased to 2.5 times more when the GPU core number increases by 4 times. Under correct parameter control, the results and precision are exactly the same as those based on GPU and CPU. Therefore, using GPU-based TeraChem software, chemists can perform large-scale ab initio quantum chemical calculation in a laboratory instead of a super-computing center.
出处
《计算机与应用化学》
CAS
2016年第5期575-581,共7页
Computers and Applied Chemistry
基金
贵州省自然科学基金资助项目(20082116)
