Mahmoodi Darian H. Investigation of Iterative Solvers Parallelization for Systems of Linear Equations Resulting from the Poisson Equation. Modares Mechanical Engineering 2023; 23 (11) :627-639
URL:
http://mme.modares.ac.ir/article-15-69692-en.html
University of Tehran , hmahmoodi@ut.ac.ir
Abstract: (478 Views)
In the present article, a survey is carried out on the parallelization of several iterative solvers of the system of linear equations resulting from the discretization of the Poisson equation using the finite difference method. In particular, the point and line Gauss-Seidel successive over-relaxation methods, as well as the conjugate gradient and stabilized biconjugate gradient methods are investigated. For the over-relaxation methods, the optimum over-relaxation coefficient is used. The parallelization is first carried out on a multi-core central processor using C++ programming language and the OpenMP library, and then for a graphics processing unit using CUDA programming language. The results show, for both the two-dimensional and three-dimensional equations, the conjugate gradient methods due to a smaller number of iterations, have less computation time. Comparing the execution time of the different methods shows that for an 8-core processing, speedups of about 10 and 5 are achieved for the two- and three-dimensional equations, respectively. Furthermore, using a graphics processing unit leads to speedups between 5 and 10 in comparison to the 8-core processing.
Article Type:
Original Research |
Subject:
Computational Fluid Dynamic (CFD) Received: 2023/06/9 | Accepted: 2023/12/11 | Published: 2023/11/1