Computational Science Technical Note CSTN-206

CSTN Home

Simulating and Visualising Lattice Boltzmann Systems with Graphical Processing Units

M. G. B. Johnson and K. A. Hawick

Archived: 2013

Abstract

Lattice Boltzmann (LB) simulation methods provide a particulate approach to simulating complex fluids and flow systems. Pseudo particles propagate and collide over a discrete mesh using localised simulation rules. Due to this locality, lattice Boltzmann methods can be readily parallelised using data-parallel decomposition methods. The performance gains permit effective simulation of large system sizes which in turn support exploration of scaling and other many length scaled phenomena that could not be explored in small model systems. LB methods also support complex flow boundaries and other problem geometries from realistic flow barriers that are hard to model properly using more conventional computational fluid flow techniques. Graphical Processing Units (GPUs) offer a relatively cheap but powerful platform for data-parallel problem decomposition. Recent developments to NVidia's Compute Unified Device Architecture (CUDA) programming language now support better dynamic memory management and GPU kernels that call one another. We discuss various GPU programming issues and performance tradeoffs in attaining high performance LB simulations in interactive time for rendered systems and very large systems sizes for non interactive cases.

Keywords: Lattice Boltzmann model; GPU; data parallelism; fluid simulation

Full Document Text: Not yet available.

Citation Information: BiBTeX database for CSTN Notes.

BiBTeX reference:

@TECHREPORT{CSTN-206,
        author = {M. G. B. Johnson and K. A. Hawick},
        title = {Simulating and Visualising Lattice Boltzmann Systems with Graphical
                Processing Units},
        institution = {Computer Science, Massey University, Auckland, New Zealand},
        year = {2013},
        number = {CSTN-206},
        keywords = {Lattice Boltzmann model; GPU; data parallelism; fluid simulation},
        owner = {kahawick},
        timestamp = {2013.09.06}
}


[ CSTN Index | CSTN BiBTeX ]