Developing a CUDA solver for large sparse matrices for MARIN
Faculty of Electrical Engineering, Mathematics and Computer Science, Delft Institute of Applied Mathematics, Delft University of Technology
Delft University of Technology, 2012
@phdthesis{de2012developing,
title={Developing a CUDA solver for large sparse matrices for MARIN},
author={DE JONG, M.},
year={2012},
school={Delft University of Technology}
}
This masters thesis has been written for the degree of Master of Science in Applied Mathematics at the faculty of Electrical Engineering, Mathematics and Computer Sciences of Delft University of Technology. The report ends a nine month internship carried out at Maritime Research Institute Netherlands (MARIN). MARIN supplies innovative products for the offshore industry and shipping companies. One of their products is a full-scale bridge simulator which can be used to train captains, steersmen and other ship workers, but also for research and consultancy purposes. Also much smaller real-time simulators are available in the form of software that can be installed and run on a single or multiple desktop PCs. To bring the simulator to a next level a new wave model is under development. The project that deals with this new wave model is called the "Interactive Waves" project. As the name explains, the project focusses on developing a simulator in which ships and waves interact. The new wave model is the Variational Boussinesq model (VBM) as suggested by Gert Klopman. However, this new realistic model brings much more computation effort with it. The VBM namely requires a solver that solves each frame (20 fps) a system S psi = b. Using VBM most of the computational time is absorbed by the solver, which makes it impossible to simulate domains larger than 200 x 400 nodes in real-time. The focus of the Master’s project was thus on developing a fast CUDA solver that would deliver a really good speed-up, so that the solving the system S psi = b is no longer the time bottleneck in the simulator.
March 16, 2012 by hgpu