Research On The Algorithm Of Charge Distribution In Molecular Dynamics Simulations Based On GPU

Because of unique hardware architecture of graphics processor unit (GPU), hardware performance improvement has been very rapid. GPU demonstrates strong performance in intensive operations, so it gets more and more attention in the field of general purpose computing. The computing concept of GPU was presented in the very early, but it has not been widely used subject to the programming complexity and accuracy issues. With the advent of the CPU performance bottlenecks, as well as modern GPU programmability on continuous improvement, the GPU intensive computing coprocessor has been widely used. Major GPU manufacturers, NVIDIA and AMD have launched their own GPGPU programming model, and continue to provide improvements. Compute unified device architecture is provided by NVIDIA for GPU’s parallel general-purpose computing architectures, and provides supporting the use of C language programming model for developed and solves using the GPU development requires excessive reliance on the shortcomings of the graphics processing API, greatly reduces the developer-oriented GPU programming difficulty, and adds support for double precision, provides a complete GPU general computing solutions.Molecular dynamics simulation is a kind of computer simulation method which apply molecular dynamics method to microstructure and movement of condensed matter for numerical simulation. The mathematical model of the molecular system is constructed by molecular dynamics, and imitates the behavior of a molecular system through calculation.Molecular dynamics simulation depends on large-scale high density calculation in many ways, and have strong demand on high performance computing ability. Combining CUDA with molecular dynamics simulation has extensive practical significance, making use of the GPU to speed up the charge distribution calculation.In the Atom bond Electronegativity Equalization Method, the ABEEMσπ Model puts forward the new molecular dynamics simulation of charge distribution calculation method based on density functional theory and the negative equilibrium theory. This model improves the atom-centered fixed charge model and has full consideration to the existence of the chemical bonds in the molecules, the application of the model in the chemical macromolecules and biological molecular system has good precision and causes some attention, its effect also got the general approved.This article is based on the MPI program of ABEEM σπ charge distribution calculation, comparing the three different parallel task partitioning schemes, and then puts forward the ABEEM σπ charge distribution MPI+CUDA parallel computing solutions. According to the MPI+CUDA parallel structure characteristics, task parallel is added to the charge distribution algorithm based on the data parallel, to improve parallelism of the whole task. The communication mechanism between GPU and CPU is improved through the overlay method between calculation and transmission, which hid communication cost and improve the efficiency of the algorithm. The actual experimental results show that CUDA can efficiently expedite speed of the charge distribution calculation.