Lattice Boltzmann for Large-Scale GPU Systems

Alan Gray; Alistair Hart; Alan Richardson; Kevin Stratford

doi:10.3233/978-1-61499-041-3-167

Lattice Boltzmann for Large-Scale GPU Systems

Alan Gray^*, Alistair Hart, Alan Richardson, Kevin Stratford

^*Corresponding author for this work

School of Physics and Astronomy

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract / Description of output

We describe the enablement of the Ludwig lattice Boltzmann parallel fluid dynamics application, designed specifically for complex problems, for massively parallel GPU-accelerated architectures. NVIDIA CUDA is introduced into the existing C/MPI framework, and we have given careful consideration to maintainability in addition to performance. Significant performance gains are realised on each GPU through restructuring of the data layout to allow memory coalescing and the adaptation of key loops to reduce off-chip memory accesses. The halo-swap communication phase has been designed to efficiently utilise many GPUs in parallel: included is the overlapping of several stages using CUDA stream functionality. The new GPU adaptation is seen to retain the good scaling behaviour of the original CPU code, and scales well up to 256 NVIDIA Fermi GPUs (the largest resource tested). The performance on the NVIDIA Fermi GPU is observed to be up to a factor of 4 greater than the (12-core) AMD Magny-Cours CPU (with all cores utilised) for a binary fluid benchmark.

Original language	English
Title of host publication	APPLICATIONS, TOOLS AND TECHNIQUES ON THE ROAD TO EXASCALE COMPUTING
Editors	K DeBosschere, EH DHollander, GR Joubert, D Padua, F Peters
Place of Publication	AMSTERDAM
Publisher	IOS Press
Pages	167-174
Number of pages	8
ISBN (Print)	978-1-61499-040-6
DOIs	https://doi.org/10.3233/978-1-61499-041-3-167
Publication status	Published - 2012
Event	14th Biennial ParCo Conference (ParCo) - Ghent, Belgium Duration: 31 Aug 2011 → 3 Sept 2011

Publication series

Name	Advances in Parallel Computing
Publisher	IOS PRESS
Volume	22
ISSN (Print)	0927-5452

Conference

Conference	14th Biennial ParCo Conference (ParCo)
Country/Territory	Belgium
Period	31/08/11 → 3/09/11

Keywords / Materials (for Non-textual outputs)

GPU
Lattice Boltzmann
CUDA
MPI
Optimisation
FLUIDS

Access to Document

10.3233/978-1-61499-041-3-167

Cite this

@inproceedings{bfd6b9c81b4d47ce8fa5e97992f49c44,

title = "Lattice Boltzmann for Large-Scale GPU Systems",

abstract = "We describe the enablement of the Ludwig lattice Boltzmann parallel fluid dynamics application, designed specifically for complex problems, for massively parallel GPU-accelerated architectures. NVIDIA CUDA is introduced into the existing C/MPI framework, and we have given careful consideration to maintainability in addition to performance. Significant performance gains are realised on each GPU through restructuring of the data layout to allow memory coalescing and the adaptation of key loops to reduce off-chip memory accesses. The halo-swap communication phase has been designed to efficiently utilise many GPUs in parallel: included is the overlapping of several stages using CUDA stream functionality. The new GPU adaptation is seen to retain the good scaling behaviour of the original CPU code, and scales well up to 256 NVIDIA Fermi GPUs (the largest resource tested). The performance on the NVIDIA Fermi GPU is observed to be up to a factor of 4 greater than the (12-core) AMD Magny-Cours CPU (with all cores utilised) for a binary fluid benchmark.",

keywords = "GPU, Lattice Boltzmann, CUDA, MPI, Optimisation, FLUIDS",

author = "Alan Gray and Alistair Hart and Alan Richardson and Kevin Stratford",

year = "2012",

doi = "10.3233/978-1-61499-041-3-167",

language = "English",

isbn = "978-1-61499-040-6",

series = "Advances in Parallel Computing",

publisher = "IOS Press",

pages = "167--174",

editor = "K DeBosschere and EH DHollander and GR Joubert and D Padua and F Peters",

booktitle = "APPLICATIONS, TOOLS AND TECHNIQUES ON THE ROAD TO EXASCALE COMPUTING",

note = "14th Biennial ParCo Conference (ParCo) ; Conference date: 31-08-2011 Through 03-09-2011",

}

Gray, A, Hart, A, Richardson, A & Stratford, K 2012, Lattice Boltzmann for Large-Scale GPU Systems. in K DeBosschere, EH DHollander, GR Joubert, D Padua & F Peters (eds), APPLICATIONS, TOOLS AND TECHNIQUES ON THE ROAD TO EXASCALE COMPUTING. Advances in Parallel Computing, vol. 22, IOS Press, AMSTERDAM, pp. 167-174, 14th Biennial ParCo Conference (ParCo), Belgium, 31/08/11. https://doi.org/10.3233/978-1-61499-041-3-167

Lattice Boltzmann for Large-Scale GPU Systems. / Gray, Alan; Hart, Alistair; Richardson, Alan et al.
APPLICATIONS, TOOLS AND TECHNIQUES ON THE ROAD TO EXASCALE COMPUTING. ed. / K DeBosschere; EH DHollander; GR Joubert; D Padua; F Peters. AMSTERDAM: IOS Press, 2012. p. 167-174 (Advances in Parallel Computing; Vol. 22).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Lattice Boltzmann for Large-Scale GPU Systems

AU - Gray, Alan

AU - Hart, Alistair

AU - Richardson, Alan

AU - Stratford, Kevin

PY - 2012

Y1 - 2012

N2 - We describe the enablement of the Ludwig lattice Boltzmann parallel fluid dynamics application, designed specifically for complex problems, for massively parallel GPU-accelerated architectures. NVIDIA CUDA is introduced into the existing C/MPI framework, and we have given careful consideration to maintainability in addition to performance. Significant performance gains are realised on each GPU through restructuring of the data layout to allow memory coalescing and the adaptation of key loops to reduce off-chip memory accesses. The halo-swap communication phase has been designed to efficiently utilise many GPUs in parallel: included is the overlapping of several stages using CUDA stream functionality. The new GPU adaptation is seen to retain the good scaling behaviour of the original CPU code, and scales well up to 256 NVIDIA Fermi GPUs (the largest resource tested). The performance on the NVIDIA Fermi GPU is observed to be up to a factor of 4 greater than the (12-core) AMD Magny-Cours CPU (with all cores utilised) for a binary fluid benchmark.

AB - We describe the enablement of the Ludwig lattice Boltzmann parallel fluid dynamics application, designed specifically for complex problems, for massively parallel GPU-accelerated architectures. NVIDIA CUDA is introduced into the existing C/MPI framework, and we have given careful consideration to maintainability in addition to performance. Significant performance gains are realised on each GPU through restructuring of the data layout to allow memory coalescing and the adaptation of key loops to reduce off-chip memory accesses. The halo-swap communication phase has been designed to efficiently utilise many GPUs in parallel: included is the overlapping of several stages using CUDA stream functionality. The new GPU adaptation is seen to retain the good scaling behaviour of the original CPU code, and scales well up to 256 NVIDIA Fermi GPUs (the largest resource tested). The performance on the NVIDIA Fermi GPU is observed to be up to a factor of 4 greater than the (12-core) AMD Magny-Cours CPU (with all cores utilised) for a binary fluid benchmark.

KW - GPU

KW - Lattice Boltzmann

KW - CUDA

KW - MPI

KW - Optimisation

KW - FLUIDS

U2 - 10.3233/978-1-61499-041-3-167

DO - 10.3233/978-1-61499-041-3-167

M3 - Conference contribution

SN - 978-1-61499-040-6

T3 - Advances in Parallel Computing

SP - 167

EP - 174

BT - APPLICATIONS, TOOLS AND TECHNIQUES ON THE ROAD TO EXASCALE COMPUTING

A2 - DeBosschere, K

A2 - DHollander, EH

A2 - Joubert, GR

A2 - Padua, D

A2 - Peters, F

PB - IOS Press

CY - AMSTERDAM

T2 - 14th Biennial ParCo Conference (ParCo)

Y2 - 31 August 2011 through 3 September 2011

ER -

Lattice Boltzmann for Large-Scale GPU Systems

Abstract / Description of output

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this