The fat-link computation on large GPU clusters for lattice QCD

Guochun Shi; Ronald Babich; Michael A. Clark; Bálint Joó; Steven Gottlieb; Volodymyr Kindratenko

doi:10.1109/SAAHPC.2012.10

The fat-link computation on large GPU clusters for lattice QCD

Guochun Shi, Ronald Babich, Michael A. Clark, Bálint Joó, Steven Gottlieb, Volodymyr Kindratenko

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

Abstract

Graphics Processing Units (GPU) are becoming increasingly popular in high performance computing due to their high performance, high power efficiency and low cost. In this paper, we present results of an effort to implement the fatlink computation - an important component of many lattice quantum chromodynamics (LQCD) calculations - on GPU clusters using the QUDA framework. Two implementations, one similar to the original CPU algorithm in the MILC code and one based on the idea of reduced communication by redundant computations, are presented and their relative advantages are discussed. In strong-scaling tests on up to 384 GPUs on Longhorn and 256 GPUs on Keeneland GPU clusters, where the CPU core to GPU ratio is 4:1 in both clusters, we achieved up to 11.4x and 8.7x node speedup when running on the two GPU clusters, respectively.

Original language	English
Title of host publication	Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012
Pages	1-10
Number of pages	10
DOIs	https://doi.org/10.1109/SAAHPC.2012.10
State	Published - 2012
Externally published	Yes
Event	2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012 - Argonne, IL, United States Duration: Jul 10 2012 → Jul 11 2012

Publication series

Name	Symposium on Application Accelerators in High-Performance Computing
ISSN (Print)	2166-5133
ISSN (Electronic)	2166-515X

Conference

Conference	2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012
Country/Territory	United States
City	Argonne, IL
Period	07/10/12 → 07/11/12

Keywords

CUDA
GPU
Lattice QCD
MILC
QUDA
Quantum chromodynamics

Access to Document

10.1109/SAAHPC.2012.10

Cite this

Shi, G., Babich, R., Clark, M. A., Joó, B., Gottlieb, S., & Kindratenko, V. (2012). The fat-link computation on large GPU clusters for lattice QCD. In Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012 (pp. 1-10). Article 6319185 (Symposium on Application Accelerators in High-Performance Computing). https://doi.org/10.1109/SAAHPC.2012.10

@inproceedings{6e8cfd6b68fc4d6c958567a3bb55f6b4,

title = "The fat-link computation on large GPU clusters for lattice QCD",

abstract = "Graphics Processing Units (GPU) are becoming increasingly popular in high performance computing due to their high performance, high power efficiency and low cost. In this paper, we present results of an effort to implement the fatlink computation - an important component of many lattice quantum chromodynamics (LQCD) calculations - on GPU clusters using the QUDA framework. Two implementations, one similar to the original CPU algorithm in the MILC code and one based on the idea of reduced communication by redundant computations, are presented and their relative advantages are discussed. In strong-scaling tests on up to 384 GPUs on Longhorn and 256 GPUs on Keeneland GPU clusters, where the CPU core to GPU ratio is 4:1 in both clusters, we achieved up to 11.4x and 8.7x node speedup when running on the two GPU clusters, respectively.",

keywords = "CUDA, GPU, Lattice QCD, MILC, QUDA, Quantum chromodynamics",

author = "Guochun Shi and Ronald Babich and Clark, {Michael A.} and B{\'a}lint Jo{\'o} and Steven Gottlieb and Volodymyr Kindratenko",

year = "2012",

doi = "10.1109/SAAHPC.2012.10",

language = "English",

isbn = "9780769548388",

series = "Symposium on Application Accelerators in High-Performance Computing",

pages = "1--10",

booktitle = "Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012",

note = "2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012 ; Conference date: 10-07-2012 Through 11-07-2012",

}

Shi, G, Babich, R, Clark, MA, Joó, B, Gottlieb, S & Kindratenko, V 2012, The fat-link computation on large GPU clusters for lattice QCD. in Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012., 6319185, Symposium on Application Accelerators in High-Performance Computing, pp. 1-10, 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012, Argonne, IL, United States, 07/10/12. https://doi.org/10.1109/SAAHPC.2012.10

The fat-link computation on large GPU clusters for lattice QCD. / Shi, Guochun; Babich, Ronald; Clark, Michael A. et al.
Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012. 2012. p. 1-10 6319185 (Symposium on Application Accelerators in High-Performance Computing).

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

TY - GEN

T1 - The fat-link computation on large GPU clusters for lattice QCD

AU - Shi, Guochun

AU - Babich, Ronald

AU - Clark, Michael A.

AU - Joó, Bálint

AU - Gottlieb, Steven

AU - Kindratenko, Volodymyr

PY - 2012

Y1 - 2012

N2 - Graphics Processing Units (GPU) are becoming increasingly popular in high performance computing due to their high performance, high power efficiency and low cost. In this paper, we present results of an effort to implement the fatlink computation - an important component of many lattice quantum chromodynamics (LQCD) calculations - on GPU clusters using the QUDA framework. Two implementations, one similar to the original CPU algorithm in the MILC code and one based on the idea of reduced communication by redundant computations, are presented and their relative advantages are discussed. In strong-scaling tests on up to 384 GPUs on Longhorn and 256 GPUs on Keeneland GPU clusters, where the CPU core to GPU ratio is 4:1 in both clusters, we achieved up to 11.4x and 8.7x node speedup when running on the two GPU clusters, respectively.

AB - Graphics Processing Units (GPU) are becoming increasingly popular in high performance computing due to their high performance, high power efficiency and low cost. In this paper, we present results of an effort to implement the fatlink computation - an important component of many lattice quantum chromodynamics (LQCD) calculations - on GPU clusters using the QUDA framework. Two implementations, one similar to the original CPU algorithm in the MILC code and one based on the idea of reduced communication by redundant computations, are presented and their relative advantages are discussed. In strong-scaling tests on up to 384 GPUs on Longhorn and 256 GPUs on Keeneland GPU clusters, where the CPU core to GPU ratio is 4:1 in both clusters, we achieved up to 11.4x and 8.7x node speedup when running on the two GPU clusters, respectively.

KW - CUDA

KW - GPU

KW - Lattice QCD

KW - MILC

KW - QUDA

KW - Quantum chromodynamics

UR - http://www.scopus.com/inward/record.url?scp=84870692776&partnerID=8YFLogxK

U2 - 10.1109/SAAHPC.2012.10

DO - 10.1109/SAAHPC.2012.10

M3 - Conference contribution

AN - SCOPUS:84870692776

SN - 9780769548388

T3 - Symposium on Application Accelerators in High-Performance Computing

SP - 1

EP - 10

BT - Proceedings - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012

T2 - 2012 Symposium on Application Accelerators in High Performance Computing, SAAHPC 2012

Y2 - 10 July 2012 through 11 July 2012

ER -

The fat-link computation on large GPU clusters for lattice QCD

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this