Study On The Optimization Method Of Molecular Dynamics Simulation Based On The Flow Processor

Molecular dynamics simulation gets macroscopic physical property by the method of simulated molecular trajectories, the compute capacity is very large. When the number of particles increases, the computation time will grow exponentially.In order to make the calculation more efficient and calculation time less,it use the powerful floating-point computing power and memory bandwidth of the GPU to solve the problem.This paper describes the powerful floating point computing power and memory bandwidth of GPU.and the CUDA programming model, software architecture, memory model, optimization of the relevant content.Based on the principle and the calculation process of molecular dynamics simulations,the paper analyses the important steps in the respect of compute capacity and the computational complexity.Finally, according to the calculation method, the paper finds out calculation of the force, the motion equations, equilibrium appropriate or need to step up to achieve in the stream processorsThe three links is achieved in the stream processors in the paper.In process of the data analysis, the efficiency based on of the stream processors is far less than the CPU efficiency under the same conditions,the main reason is a thread for all particles, coupled with the data in main memory and memory copy overhead so that the total time lengthened. In the optimization process,aim at the question of the parallelization of particles, it improves the number of threads by setting the kerne1<<<grid, block>>>0in the grid and block,and collaborative particle correlation data in each thread in parallel to give full play to the highly parallel processing capabilities of the stream processors.On the other hand, in the process of particle calculation frequently read data in global memory and shared memory,The data address alignment and bank conflict will lead to the number of memory access change eventually fetch longer,the proposed merger of access and data offset method to solve.Finally, there is a analysis of the achieve of the CPU-based and stream-based processor,In three steps based on the stream processors to optimize the former efficiency is not the CPU, and optimized efficiency far more than the CPU,efficiency of the CPU4to5times,1.5to2times,1.5to2times respectively.Therefore, based on the stream processors optimized for molecular dynamics simulation is applicable. With the number of stream processors and memory size to improve efficiency will have a greater improvement.