The Study Of Nonbonded Interaction Calculation Of Molecular Dynamics Based On CUDA

Molecular dynamics is a main calculation method in the atomic level in solid and liquid material. It is for the study of material microstructure, thermodynamic properties and bal-ance transport properties and so on, and widely applied in physical science, material science, life sciences and so on. In the process of computation, only the particle number is infinite, can reproduce the real physical macro behavior, makes the molecular dynamics calculation need spend too much time and general with weeks for the unit. In a molecular dynamics sim-ulation, the nonbonded interaction calculation is the highest compute-intensive calculation in the process of the whole simulation calculation. Therefore, the search of the nonbonded interaction calculation method of the molecular dynamics simulation is of great significance for realizing complex system. In the past two years, the CUDA technology is launched, it is a new software and hardware structure, and contributed to the solution of problems complex calculations in science and engineering. In this paper, we will discuss the calculation and accelerate scheme of the nonbonded interaction, then design and implement the molecular dynamics in the workbench of CUDA. The specific work includes the following aspects:First of all, analyzes the main characteristics of molecular dynamics simulation method, and discusses the feasibility and superiority of CUDA technology to improve the efficiency.Then, the article analyzes and discusses the current acceleration scheme of molecular dynamics simulation. We will discuss the different parallel schemes in different workbench, based on cluster and CUDA platform. Then study the three famous parallel algorithms, atomic decomposition, force decomposition, space decomposition and their combination.Again, design a nonbonded interaction calculation method of solute and solute interac-tion, solvent and solvent interaction, solute and solvent interaction based on CUDA. Porting the nonbonded interaction computing tasks to GPUs, make full use of the parallel computing advantage on GPU, improve the computational performance.Finally, this paper discusses the CPU-GPU hybrid platform and GPUs scheduling method, used many GPUs device for molecular dynamics simulation, make full use of the computing resources in the system, optimized the simulation accuracy and efficiency. In this paper, the CUDA platform is effectively applied in molecular dynamics simula-tion field, improving the efficiency of molecular dynamics simulation. Especially, the calcu-lation of the nonbonded interaction is assigned to GPUs, making the efficiency greatly raised. The use of CUDA platform provides a new idea for the research of biological molecules sim-ulation, also have the certain reference value.