Multi-GPUs Parallel Computing And Its Application In Particle Simulation

Particle-in-cell( PIC) method is an important method to research the motion of charged particles. Along with the rapid development of computer technology, the PIC method has been favored by many domestic and international research institutions since its birth. The application field of PIC is very wide, and it has become an important research means. But with the development of scientific computing, computational scale increases rapidly. The traditional computer based on central processing unit(CPU) has been unable to meet the computational requirements, subjected to physical structure and energy consumption. GPU is different from CPU; it adopts a many-core microprocessor system structure, and has hundreds or even thousands of Steam Processor. Each stream processor has a fully pipelined integer arithmetic logic unit( ALU) and floating point unit( FPU), just like a single core CPU, which makes the GPU has powerful computing ability. With the rapid development of GPU, its application extends to the field of High Performance Computing(HPC). How to coordinate CPU and GPU together and play to their own excellent computing power, has already become the key point in the field of HPC. How to design excellent algorithms around the GPU accelerated implementation has gradually become an important development trend in the HPC field.This paper mainly researches parallel computing technology of Multi-GPUs in personal computer, and completes the interaction of laser and plasma in parallel based on Multi-GPUs. This paper mainly studies the application advantages of different computer architectures designed around GPU, especially studies the application of the different computing system architecture in particle parallel simulation. The paper also describes the latest PIC research results and development trend on the GPU technology both in domestic and international research institutions. The paper further summarizes the CUDA’s technology points and application advantages in different GPU computing system. At last, the paper summarizes and refines Multi-GPUs parallel algorithm, based on relativistic laser plasma interactions simulation software LPICMCC++ which contains collision effects developed by UESTC, laser plasma interaction mathematical model and serial algorithm of LPICMCC++. The paper adopts the industry’s latest technology Uuified Memory, rewrites the program source code of different modules in the LPICMCC++ solvers, transplantes the rewriting code in parallel computing environment to speed up the calculation.At last, the paper verifies the whole physical process of laser and plasma interaction in Multi-GPUs engineering environment and compares the two results calculated by serial LPICMCC++ code and parallel LPICMCC++ code to verify the correction of Multi-GPUs parallel code. By calculating the time and the results of different particle states, the paper summarizes the speedup of Multi-GPUs parallel implementation, and explaines the reason of the error in the result.