Multi-GPU Acceleration of Algebraic Multi-Grid Preconditioners for Elliptic Field Problems

Author

Richter, Christian ; Schops, Sebastian ; Clemens, Markus

Author_Institution

Dept. of Electromagn. Theor., Bergische Univ. Wuppertal, Wuppertal, Germany

Volume

51

Issue

3

fYear

2015

fDate

Mar-15

Firstpage

1

Lastpage

4

Abstract

In this contribution, a multi-graphic processing unit (GPU) implementation of Krylov sub-space methods with algebraic multi-grid preconditioners is proposed. It is used to solve large linear systems stemming from finite element or finite difference discretizations of elliptic problems as they occur, e.g., in electrostatics. The distribution of data across multiple GPUs and the effects on memory and speed are discussed when using an approach that preserves the effects of fine-grained parallelism with shared memory on the GPU while distributing data across multiple GPUs with minimal communication effort.

Keywords

electromagnetic field theory; finite element analysis; graphics processing units; partial differential equations; GPU; Krylov subspace method; algebraic multigrid preconditioners; elliptic field problems; fine-grained parallelism; finite difference discretization; finite element method; linear system; multiGPU acceleration; multigraphic processing unit; partial differential equations; Acceleration; Electromagnetics; Graphics processing units; Libraries; Linear systems; Mathematical model; Sparse matrices; Algebraic multi-grid (AMG); CUDA; conjugate gradients (CGs); multi-graphic processing unit (GPU); sparse matrix vector multilication (SpMV);

fLanguage

English

Journal_Title

Magnetics, IEEE Transactions on

Publisher

ieee

ISSN

0018-9464

Type

jour

DOI

10.1109/TMAG.2014.2357332

Filename

7093525