Abstract: In particle methods, the application of neighbor particle search algorithm can quickly get the information of neighbor particle, but the tradition neighbor particle search puts a high challenge to computing speed from large-scale particle system data calculation. The compute capability of GPU and develop environment of CUDA are studied. Based on parallel multithread processing technology of GPU（Graphic Processing Unit）, a parallel search algorithm is proposed. The result shows that the algorithm of parallel neighbor particle search based on GPU can accelerate the process of neighbor particle search and reduce the time significantly, and get the acceleration of more than 290 times, and shows high-performance processing power in large-scale particle system.

Key words: Compute Unified Device Architecture（CUDA）, Graphic Processing Unit（GPU）, particle method, neighbor particle search

摘要： 在粒子方法中，运用邻近粒子搜索算法可以快速获取每个粒子的邻近粒子信息。由于粒子方法模拟一个体系的行为所采用的粒子数据是十分庞大的，对计算机的运算速度提出了挑战。研究了GPU的计算能力和CUDA开发环境，利用GPU的并行多线程处理技术，提出了一种并行邻近粒子搜索算法。实验结果表明，基于CUDA的并行邻近粒子搜索算法，加快了邻近粒子搜索过程，显著地减少了计算时间，成功实现了硬件加速，可获取290以上的加速比，对大规模粒子系统呈现出高效的处理能力。

关键词: 统一计算设备框架（CUDA）, 图形处理单元（GPU）, 粒子方法, 邻近粒子搜索