GPU-base (Graphics Processing Unit-based) multi-particle transport simulation method

Abstract

The invention discloses a GPU-base (Graphics Processing Unit-based) multi-particle transport simulation method, which comprises the following steps that initiation is carried out; the thread number of a GPU is allocated according to the number of particles; the next collision points of the particles are computed in parallel by the GPU, whether the weight splitting or death of the particles takes place is judged, and events which are about to take place at the collision points are given; a computation result is returned to a CPU (central processing unit), and the CPU reallocates the thread number of the GPU according to the number and types of the events; the events are processed in parallel in the GPU, and the new energy, azimuths and other information of the particles and the information of produced secondary particles after the occurrence of the events are given and returned to the CPU; the CPU gathers the information and returns to the start of a program, and the steps are cycled until a simulation result of particle transport is statistically outputted. According to the embodiment of the invention, the GPU-base multi-particle transport simulation method can realize high-efficient experimental particle transport simulation, and while energy consumption is reduced, the efficiency of computation is increased.

Description

technical field [0001] Exemplary embodiments of the present invention generally relate to the technical field of multi-particle transport simulation, and in particular, relate to a GPU-based Monte Carlo transport method applicable to neutron, photon and electron mixed transport research. Background technique [0002] Particle transport simulation is of great significance in scientific and engineering calculations. So far, its solution algorithms can be divided into deterministic methods and Monte Carlo methods. Since the Monte Carlo method has the characteristics that the convergence speed has nothing to do with the dimension of the problem, is less affected by the conditional constraints of the problem, and has a simple program structure, when dealing with particle transport, the Monte Carlo method is more effective in dealing with cross-section and energy, and scattering anisotropy. It has unique advantages when dealing with problems such as heterogeneity, medium inhomogen...

AI Technical Summary

Problems solved by technology

But its biggest disadvantage is that the convergence speed is slow, and the probabilistic nature of the error requires a large number of samples to ensure accuracy.

At present, the realization of particle transport simulation calculation is mainly based on CPU. Since the insufficient computing power of CPU directly affects its calculation efficiency, it can only calculate a small space size and time size within an acceptable computer time; or in order to increase the amount of calculation Large-scale or even ultra-large-scale CPU computer clusters can be built, but this will cause problems such as excessive power consumption, high use and maintenance costs, etc.

Images