aalto1 untyped-item.component.html

Texture-based GPU parallelization of isogeometric analysis

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Engineering | Bachelor's thesis

Department

Mcode

Language

en

Pages

34

Series

Abstract

Isogeometric analysis offers an approach to finite element analysis which uses non-rational uniform b-splines as basis functions. While isogeometric finite element methods have the potential to offer greater accuracy and robustness over their classical counterparts, these benefits come at a cost during the integration process. This thesis aims to reduce the total computational time through a novel approach to parallelizing the isogeometric finite element analysis process. A texture-based parallelization approach is developed and used to simulate the Cahn-Hilliard equation in one dimension. Furthermore, the technique parallelizes the whole analysis process, thereby minimizing any communication overhead arising from transferring data between the CPU and GPU during runtime. Lastly, the results of the parallel implementation are compared to other approaches for simulating the Cahn-Hilliard equation found in literature.

Description

Supervisor

St-Pierre, Luc

Thesis advisor

Khakalo, Sergei
Kranzosch, Ole

Other note

Citation

Endorsement

Review

Supplemented By

Referenced By