Implementation of the Deflated Preconditioned Conjugate Gradient Method for Bubbly Flow on the Graphical Processing Unit (GPU)

Master thesis (2010)

Authors

Contributors

C. Vuik (mentor)
Programme
Numerical Analysis (EEMCS) (TU Delft)
Copyright: © 2010 Gupta, R.
More Info
expand_more

Published Date

23-08-2010

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Programme

Numerical Analysis

Abstract

In this work we have implemented the Iterative Method of Conjugate Gradients with two levels of Preconditioning to solve a System of Linear Equations on Graphical Processing Unit (GPU). This system represents the discretized Pressure equation resulting from the Level Set Method Solution of the Incompressible Navier Stokes Equation used to compute Bubbly Flows. We have tried to explore the problem space with different grid sizes, number of preconditioning blocks and deflation vectors. The results show that when the methods for preconditioning are chosen so that they can exhibit ample parallelism we can achieve considerable performance up to 20 times better than the CPU version. We show in our analysis that we are very close to maximum achievable speedup. We also report on the accuracy of our results and argue that GPUs can be beneficial in solving such problems efficiently.

Files

Thesis_rohit.pdf
(pdf | 2.49 Mb)