期刊名称：International Journal of Advanced Research In Computer Science and Software Engineering
印刷版ISSN：2277-6451
电子版ISSN：2277-128X
出版年度：2013
卷号：3
期号：10
出版社：S.S. Mishra
摘要：The Graphical Processing Unit is gaining popularity as a powerful commodity processor for computational modelling analysis in many engineering and scientific applications. Majority of scientific and engineering computational analysis codes are based on techniques such as the finite element method that results in large memory bound sparse matrix linear equation systems. The associated sparse matrices are stored in different sparse matrix compression formats in computational analysis software code developments. Solution of such large sparse matrix linear system of equations are integral and time consuming part of the computations which are solved via iterative solution approaches such as the pre-conditioned conjugate gradient method. This paper evaluates the effect of sparse matrix compression formats and the associated data locality on the performance of preconditioned conjugate gradient solvers in a Graphical Pro cessing Unit (GPU) computing system. Computational performance variations and inferences between GPU and CPU platforms are discussed in the context of a finite element based engineering application software analysis code for composite process flow modelling
