Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment

Fredrik Robertsén; Jan Westerholm; Keijo Mattila

doi:10.1109/PDP.2015.71

Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment

Fredrik Robertsén, Jan Westerholm, Keijo Mattila

Research output: Chapter in Book/Conference proceeding › Conference contribution › Scientific › peer-review

10 Citations (Scopus)

Abstract

The lattice Boltzmann method is a well-established numerical approach for complex fluid flow simulations. Recently general-purpose graphics processing units have become accessible as high-performance computing resources at large-scale. We report on implementing a lattice Boltzmann solver for multi-GPU systems that achieves 0.69 PFLOPS performance on 16384 GPUs. In addition to optimizing the data layout on the GPUs and eliminating the halo sites, we make use of the possibility to overlap data transfer between the host CPU and the device GPU with computing on the GPU. We simulate flow in porous media and measure both strong and weak scaling performance with the emphasis being on a large scale simulation using realistic input data.

Original language	Undefined/Unknown
Title of host publication	Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on
Editors	Daneshtalab Masoud, Aldinucci Marco, Leppänen Ville, Lilius Johan, Brorsson Mats
Publisher	IEEE
Pages	604–609
ISBN (Print)	978-1-4799-8492-3
DOIs	https://doi.org/10.1109/PDP.2015.71
Publication status	Published - 2015
MoE publication type	A4 Article in a conference publication
Event	Euromicro International Conference on Parallel, Distributed and Network-Based Processing - 23rd Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP 2015) Duration: 4 Mar 2015 → 6 Mar 2015

Conference

Conference	Euromicro International Conference on Parallel, Distributed and Network-Based Processing
Period	04/03/15 → 06/03/15

Keywords

GPU
Lattice Boltzmann methods

Access to Document

10.1109/PDP.2015.71

Cite this

Robertsén, F., Westerholm, J., & Mattila, K. (2015). Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment. In D. Masoud, A. Marco, L. Ville, L. Johan, & B. Mats (Eds.), Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on (pp. 604–609). IEEE. https://doi.org/10.1109/PDP.2015.71

Robertsén, Fredrik ; Westerholm, Jan ; Mattila, Keijo. / Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment. Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on. editor / Daneshtalab Masoud ; Aldinucci Marco ; Leppänen Ville ; Lilius Johan ; Brorsson Mats. IEEE, 2015. pp. 604–609

@inproceedings{b4152242403844bc83552f656ad89686,

title = "Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment",

abstract = "The lattice Boltzmann method is a well-established numerical approach for complex fluid flow simulations. Recently general-purpose graphics processing units have become accessible as high-performance computing resources at large-scale. We report on implementing a lattice Boltzmann solver for multi-GPU systems that achieves 0.69 PFLOPS performance on 16384 GPUs. In addition to optimizing the data layout on the GPUs and eliminating the halo sites, we make use of the possibility to overlap data transfer between the host CPU and the device GPU with computing on the GPU. We simulate flow in porous media and measure both strong and weak scaling performance with the emphasis being on a large scale simulation using realistic input data.",

keywords = "GPU, Lattice Boltzmann methods, GPU, Lattice Boltzmann methods, GPU, Lattice Boltzmann methods",

author = "Fredrik Roberts{\'e}n and Jan Westerholm and Keijo Mattila",

year = "2015",

doi = "10.1109/PDP.2015.71",

language = "Odefinierat/ok{\"a}nt",

isbn = "978-1-4799-8492-3",

pages = "604–609",

editor = "Daneshtalab Masoud and Aldinucci Marco and Lepp{\"a}nen Ville and Lilius Johan and Brorsson Mats",

booktitle = "Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on",

publisher = "IEEE",

note = "Euromicro International Conference on Parallel, Distributed and Network-Based Processing ; Conference date: 04-03-2015 Through 06-03-2015",

}

Robertsén, F, Westerholm, J & Mattila, K 2015, Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment. in D Masoud, A Marco, L Ville, L Johan & B Mats (eds), Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on. IEEE, pp. 604–609, Euromicro International Conference on Parallel, Distributed and Network-Based Processing, 04/03/15. https://doi.org/10.1109/PDP.2015.71

Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment. / Robertsén, Fredrik; Westerholm, Jan; Mattila, Keijo.
Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on. ed. / Daneshtalab Masoud; Aldinucci Marco; Leppänen Ville; Lilius Johan; Brorsson Mats. IEEE, 2015. p. 604–609.

Research output: Chapter in Book/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment

AU - Robertsén, Fredrik

AU - Westerholm, Jan

AU - Mattila, Keijo

PY - 2015

Y1 - 2015

N2 - The lattice Boltzmann method is a well-established numerical approach for complex fluid flow simulations. Recently general-purpose graphics processing units have become accessible as high-performance computing resources at large-scale. We report on implementing a lattice Boltzmann solver for multi-GPU systems that achieves 0.69 PFLOPS performance on 16384 GPUs. In addition to optimizing the data layout on the GPUs and eliminating the halo sites, we make use of the possibility to overlap data transfer between the host CPU and the device GPU with computing on the GPU. We simulate flow in porous media and measure both strong and weak scaling performance with the emphasis being on a large scale simulation using realistic input data.

AB - The lattice Boltzmann method is a well-established numerical approach for complex fluid flow simulations. Recently general-purpose graphics processing units have become accessible as high-performance computing resources at large-scale. We report on implementing a lattice Boltzmann solver for multi-GPU systems that achieves 0.69 PFLOPS performance on 16384 GPUs. In addition to optimizing the data layout on the GPUs and eliminating the halo sites, we make use of the possibility to overlap data transfer between the host CPU and the device GPU with computing on the GPU. We simulate flow in porous media and measure both strong and weak scaling performance with the emphasis being on a large scale simulation using realistic input data.

KW - GPU

KW - Lattice Boltzmann methods

KW - GPU

KW - Lattice Boltzmann methods

KW - GPU

KW - Lattice Boltzmann methods

U2 - 10.1109/PDP.2015.71

DO - 10.1109/PDP.2015.71

M3 - Konferensbidrag

SN - 978-1-4799-8492-3

SP - 604

EP - 609

BT - Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on

A2 - Masoud, Daneshtalab

A2 - Marco, Aldinucci

A2 - Ville, Leppänen

A2 - Johan, Lilius

A2 - Mats, Brorsson

PB - IEEE

T2 - Euromicro International Conference on Parallel, Distributed and Network-Based Processing

Y2 - 4 March 2015 through 6 March 2015

ER -

Robertsén F, Westerholm J, Mattila K. Lattice Boltzmann Simulations at Petascale on Multi-GPU Systems with Asynchronous Data Transfer and Strictly Enforced Memory Read Alignment. In Masoud D, Marco A, Ville L, Johan L, Mats B, editors, Parallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on. IEEE. 2015. p. 604–609 doi: 10.1109/PDP.2015.71