Developing a CUDA solver for large sparse matrices for MARIN

Master thesis (2012)

Authors

M.A. De Jong

Contributors

C. Vuik (mentor)

H.X. Lin (mentor)

A. Van der Ploeg (mentor)

A. Ditzel (mentor)

Department

Numerical Analysis () (TU Delft)

MARIN Parallelism CUDA Conjugate Gradient Parallel Poisson Multigrid GPU Preconditioning CG RRB Repeated Red-Black Incomplete LU Cholesky VBM Variational Boussinesq Diagonal scaling Sparse Linear

To reference this document use:

http://resolver.tudelft.nl/uuid:cb47a416-9547-4e91-81d8-9c8b060ed7d8

More Info

expand_more

Published Date

31-08-2012

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.

Faculty

Electrical Engineering, Mathematics and Computer Science

Department

Numerical Analysis

Abstract

The Maritime Research Institute Netherlands (MARIN) supplies innovative products for the offshore industry and shipping companies. Among their products are highly realistic real-time bridge simulators. Currently, the waves are deterministic and are not affected by ships, moles, breakwaters, piers, or any other object. To bring the simulators to the next level, a new interactive wave model is being developed. This is the so-called Variational Boussinesq model (VBM) as proposed by Klopman. The main improvement will be that the waves and ships really interact, i.e., the movements of the ship are influenced by the waves and the waves in their turn are influenced by the ship. However, one pays for the higher realism: the new model is much more computational intensive and therefore a really fast solver is needed to fulill the requirements of real-time simulation. In this thesis we present how a very efficient iterative solver can be combined with a very efficient implementation on the graphical processing unit (GPU). Speed up factors of more than 30 can be obtained for realistic problems. With the new solver interactive waves can be computed in real-time for large domains. Necessary theory on iterative solvers and GPU programming is included.

Files

Jong_afst.pdf

(pdf | 6.31 Mb)