Developments And Applications Of GPU Computing In Quantum Chemical Calculations

In this paper, we focus on the applications and developments of GPUcomputing in computational chemistry. Firstly, we presented a newalgorithm with GPU computing in computational chemistry and theapplication of GPU based computational chemistry (quantum mechanicsand molecular mechanics) programs in recent years. Secondly, wedetailedly introduced our independent researches and development ofsemi-empirical program of GPU-based acceleration and two-electronrepulsion integral algorithm on GPU.This thesis is divided as follows:(1) Description of computationalchemistry with ab initio methods, DFT methods and molecular dynamicstheory, existing problems and performance bottlenecks of CPUprocedures.(2) The related knowledges of GPU computing and NvidiaCUDA architecture.(3) Introduction of progress of TeraChem, ACEMDand other GPU programs and our attempt to use GPU QM/MM bycombining TeraChem with Amber.(4) Semi-empirical methods are stillplaying important roles in today’s computational chemistry, by theirhundred times faster speed than common ab initio in performances. But todeal with the biological systems, which normally consist of thousands totens of thousands of atoms, the semi-empirical methods need to bespeeded up greatly and break through. So we present our GPU enabledsemi-empirical code implemented with CUDA, and focused on the bottlenecks of speed in semi-empirical calculations, matrixdiagonalizations and constructions of density matrixes. A benchmark testperformed for large molecules with our code on the GPU achieves up to30-fold speedup over a traditional CPU implementation.(5) For most abinitio methods, the first step is to calculate the large number oftwo-electron repulsion integrals, which is the major bottleneck of abinitio calculation programs. We used Nvidia CUDA platform to rewrittenthe two-electron integral calculation part of ab initio algorithms. Throughour ongoing exploration, the new code using the GPU can significantlyimprove the computing speed of two-electron integrals computation.Most of the present computational chemistry programs weredeveloped during the1970s to1980s, the old program frameworks aretheir weaknesses in the contemporary era. The recent rise of GPGPU(General Purpose Graphical Processing Units) technology with nVidia’sCUDA (Computer Unified Device Architecture) has brought gospel tocomputational chemistry. GPU computing also provides a newopportunity for computational chemistry in China, which could be a goodopportunity to develop our new computational chemistry programs whichhave Chinese intellectual property with GPU computing technology. It isthe time to change the situation that we Chinese computational chemistryscientists can only buy foreign softwares with high price.