航空学报 > 2010, Vol. 31 Issue (2): 249-256

流体力学、飞行力学与发动机

基于GPU和隐式格式的CFD并行计算方法

张兵, 韩景龙   

  1. 南京航空航天大学 航空宇航学院
  • 收稿日期:2009-06-16 修回日期:2009-09-15 出版日期:2010-02-25 发布日期:2010-02-25
  • 通讯作者: 韩景龙

Parallel Computing Methods for CFD Using a GPU and Implicit Scheme

Zhang Bing, Han Jinglong   

  1. College of Aerospace Engineering, Nanjing University of Aeronautics and Astronautics
  • Received:2009-06-16 Revised:2009-09-15 Online:2010-02-25 Published:2010-02-25
  • Contact: Han Jinglong

摘要: 从图形处理器(GPU)架构特点出发,提出了基于数据并行的隐式计算流体力学(CFD)求解方法,空间离散格式采用迎风Roe格式,计算网格适用于结构和非结构网格。采用统一计算设备架构(CUDA)技术实现了GPU上的隐式CFD并行计算。分别在Intel Core 2 Quad 3.0 GHz CPU和NVIDIA GTX280 GPU上进行了计算,结果表明隐式格式计算速度是显式格式6倍以上,采用显式格式的计算加速比达到28倍,采用隐式格式计算加速比达到了28.7倍,同时计算加速比随计算规模的增加而增加。计算结果和实验结果较为吻合。

关键词: 计算流体力学, 图形处理器, 并行计算, 隐式格式, 统一计算设备架构

Abstract: Based on the features of the graphics processing unit (GPU) architecture, an implicit data-parallel scheme is developed to solve computaitional fluid dynamics (CFD) problems. The proposed method is applicable to structured and unstructured meshes and uses an upwind scheme to achieve more accurate results. The method is implemented on NVIDIA GTX280 GPU by employing compute unifified device anchitecture (CUDA) technology and compared with Intel Core 2 Quad 3.0 GHz CPU. The results indicate that the implicit scheme proposed in this article is 6 times faster than the explicit scheme with the same hardware and the computation is sped up to 28.7 times by using the GPU and implicit scheme comparing to 28 times by using explicit scheme, and it will be more efficient for larger scale problems. Finally, the results with this scheme exhibit good agreement with existing experimental data.

Key words: computational fluid dynamics, graphics processing unit, parallel computing, implicit scheme, compute unified device architecture

中图分类号: