理論応用力学講演会 講演論文集
第58回理論応用力学講演会
セッションID: PD3-1
会議情報

パネルディスカッション3
On GPGPU Computing for Finite Element Method
*奥田 洋司
著者情報
会議録・要旨集 フリー

詳細
抄録

Solving large systems of linear equations represents the most time consuming part of Finite Element Analyses. For the case of very large matrices, Krylov solvers like Conjugate Gradients or GMRES are usually preferred. In this research we investigate the acceleration of Krylov methods on commodity graphic hardware. To assess the real-world performance, we test the performance and the cost/performance of the solvers on a large collection of test matrices. For large matrices, up to 15x speedup is obtained when using three GPUs.

著者関連情報
© 2009 社団法人土木学会
前の記事 次の記事
feedback
Top