The Research And Implementation Of Parallel Computation Of Molecular Dynamics Simulation For Liquid Metals Solidification Processes On GPU

With the development of computer science and integrated circuit technology,GPU(Graphics Processing Unit)has developed as a coprocessor of CPU to complete the large-scale computing tasks,GPU is designed for intensive and highly parallel computation.On the other hand,molecular dynamics(MD)simulation,as a kind of discrete simulation method,can be used to obtain a variety of physical properties of molecular simulation system,mainly through the qualitative and quantitative analysis to the large molecular movement.MD simulation has been widely used in physics,chemistry,biology,materials,medicine and other fields.However,a large amount of computation involved in the simulation system restrict the development of their research.Therefore,the research of molecular dynamics simulation in the liquid metals solidification processes with GPU based on CUDA(Computer Unified Device Architecture)has important theoretical and practical significance.In their processes,the simulation results will be more accurate when the number of atoms in the system is larger,and then we can obtain more microscopic properties and predict the trend of development,which can provide guidance on the actual production.This thesis introduces GPU parallel computing technology and molecular dynamics simulation method in detail,improves the traditional spatial decomposition method to fit with the GPU computing architecture and then gives a fine-grained spatial decomposition algorithm.This algorithm narrows task computing grain and makes full use of a large number of processing cores in GPU.On the basis of the fine-grained spatial decomposition method proposed in thesis,the thesis gives the parallel computing model for the most time-consuming component(update of neighbor lists and interaction force calculation)in MD simulation calculation,and then analysis the GPU-MD parallel algorithm.This algorithm is implemented on the NVIDIA Tesla M2050 GPU,enlarging the scale of the simulation system to a simulation system involving 10,000,000 atoms.In addition,a number of evaluations and tests about this parallel computing model are discussed and compared,ranging from executions on different precision enabled-CUDA versions,over various types of GPU(NVIDIA 480GTX,580GTX and M2050)to CPU clusters with different number of CPU cores.The experimental results demonstrate that GPU-based calculations are typically 9?11 times faster than the corresponding sequential execution and approximately 1.5?2 times faster than 16 CPU cores clusters implementations.On the basis of the simulated results,the comparisons between the theoretical results and the experimental ones are executed through pair distribution function,and the good agreement between the two are observed.And then this thesis adopts clusters model method to obtain more complete and larger cluster structures in the actual macroscopic materials,such as nano-crystals formed in the processes of metal solidification and different nucleation and evolution mechanism of nano-clusters with large-sized system.