Vis enkel innførsel

dc.contributor.advisorHolm, Håvardnb_NO
dc.contributor.authorThyholdt, Kristoffer Clausennb_NO
dc.date.accessioned2014-12-19T12:08:24Z
dc.date.available2014-12-19T12:08:24Z
dc.date.created2012-11-08nb_NO
dc.date.issued2012nb_NO
dc.identifier566322nb_NO
dc.identifierntnudaim:7271nb_NO
dc.identifier.urihttp://hdl.handle.net/11250/238308
dc.description.abstractThe lattice Boltzmann method has become a valuable tool in computational fluid dynamics, one of the reasons is due to the simplicity of its coding. In order to maximize the performance potential of today s computers, code has to be optimized for parallel execution. In order to achieve parallel execution of the lattice Boltzmann method, the data dependency has to be solved. And to get good performance, the memory has to be organized for unit stride access. Here we investigate the most known algorithms for lattice Boltzmann, and implement a code which runs on a parallel graphics processor, using a library for parallelization called C++ AMP. Furthermore, we show how the code compares to known solutions of fluid flows to verify the numerical results. The optimized parallel code achieves a speed up of 650 times the un-optimized code, on a current generation high-end graphics card.nb_NO
dc.languageengnb_NO
dc.publisherInstitutt for marin teknikknb_NO
dc.subjectntnudaim:7271no_NO
dc.subjectMIMART Marin teknikk (2 årig)no_NO
dc.subjectMarin hydrodynamikkno_NO
dc.titleLattice Boltzmann Simulations on a GPU: An optimization approach using C++ AMPnb_NO
dc.typeMaster thesisnb_NO
dc.source.pagenumber71nb_NO
dc.contributor.departmentNorges teknisk-naturvitenskapelige universitet, Fakultet for ingeniørvitenskap og teknologi, Institutt for marin teknikknb_NO


Tilhørende fil(er)

Thumbnail
Thumbnail
Thumbnail

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel