蔡勇, 李光耀, 王琥. 基于多重网格法和GPU并行计算的大规模壳结构快速计算方法[J]. 工程力学, 2014, 31(5): 20-26. DOI: 10.6052/j.issn.1000-4750.2012.12.0942
引用本文: 蔡勇, 李光耀, 王琥. 基于多重网格法和GPU并行计算的大规模壳结构快速计算方法[J]. 工程力学, 2014, 31(5): 20-26. DOI: 10.6052/j.issn.1000-4750.2012.12.0942
CAI Yong, LI Guang-yao, WANG Hu. A FAST CALCULATION METHOD FOR LARGE-SCALE SHELL STRUCTURE BASED ON MULTIGIRD METHOD AND GPU PARALLEL COMPUTING[J]. Engineering Mechanics, 2014, 31(5): 20-26. DOI: 10.6052/j.issn.1000-4750.2012.12.0942
Citation: CAI Yong, LI Guang-yao, WANG Hu. A FAST CALCULATION METHOD FOR LARGE-SCALE SHELL STRUCTURE BASED ON MULTIGIRD METHOD AND GPU PARALLEL COMPUTING[J]. Engineering Mechanics, 2014, 31(5): 20-26. DOI: 10.6052/j.issn.1000-4750.2012.12.0942

基于多重网格法和GPU并行计算的大规模壳结构快速计算方法

A FAST CALCULATION METHOD FOR LARGE-SCALE SHELL STRUCTURE BASED ON MULTIGIRD METHOD AND GPU PARALLEL COMPUTING

  • 摘要: 该文采用将EBE计算策略、多重网格法以及GPU并行计算方法三者相结合的计算策略, 设计了一种新颖的迭代求解方法, 可以有效的提高大规模壳结构的有限元分析效率。该方法中, EBE计算策略将总体运算分解到单元上进行, 可以节约计算内存, 提高单机上问题的求解规模, 并且可以有效地提高隐式有限元算法的并行性;多重网格法通过在疏密不同的网格层上进行迭代, 平滑不同频率的误差分量, 可以加快迭代收敛速度;GPU并行计算方法可以在较低硬件成本的前提下实现高效的并行计算。该文采用统一计算架构(Compute Unified Device Architecture, CUDA)进行程序的编制, 并在采用GTX460显卡的个人计算机执行。数值计算结果表明该方法在保证计算精度的同时可以取得较高的计算加速比。

     

    Abstract: A novel iterative solution method for implicit finite element equations based on EBE scheme, multigrid method and GPU parallel computing method is designed to speed up the finite element analysis of a large-scale shell structure. In this method, EBE calculation strategies disassemble a global computation to a local element, which can reduce memory consumption and significantly increase the solution scale, more importantly it can improve the parallelism of implicit finite element calculation. Multigrid method can accelerate the convergence of iteration by using different mesh densities to eliminate the different frequency components of errors. And, GPU parallel computing is a novel parallel approach to reduce the time of computation with lower cost. The program of this method is compiled by CUDA (Compute Unified Device Architecture) and then implemented in a personal computer with a GTX460 graphics card, the calculation results show that the method can achieved a high computing speed-up ratio with high calculation accuracy.

     

/

返回文章
返回