Strong scaling of general-purpose molecular dynamics simulations on GPUs

被引：528

作者：

Glaser, Jens ^{[1
]}

Trung Dac Nguyen ^{[3
]}

Anderson, Joshua A. ^{[1
]}

Lui, Pak ^{[4
]}

Spiga, Filippo ^{[5
]}

Millan, Jaime A. ^{[2
]}

Morse, David C. ^{[6
]}

Glotzer, Sharon C. ^{[1
,2
]}

机构：

[1] Univ Michigan, Dept Chem Engn, Ann Arbor, MI 48109 USA

[2] Univ Michigan, Dept Mat Sci & Engn, Ann Arbor, MI 48109 USA

[3] Oak Ridge Natl Lab, Natl Ctr Computat Sci, Oak Ridge, TN 37831 USA

[4] Mellanox Technol Inc, Sunnyvale, CA 94085 USA

[5] Univ Cambridge, High Performance Comp Serv, Cambridge CB2 1RX, England

[6] Dept Chem Engn & Mat Sci, Minneapolis, MN 55455 USA

来源：

COMPUTER PHYSICS COMMUNICATIONS | 2015年 / 192卷

基金：

美国国家科学基金会;

关键词：

Multi-GPU; Molecular dynamics; MPI/CUDA; Strong scaling; Weak scaling; Domain decomposition; LAMMPS; ALGORITHMS; EFFICIENT; CLUSTERS; RDMA; MPI;

D O I：

10.1016/j.cpc.2015.02.028

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

We describe a highly optimized implementation of MPI domain decomposition in a GPU-enabled, general-purpose molecular dynamics code, HOOMD-blue (Anderson and Glotzer, 2013). Our approach is inspired by a traditional CPU-based code, LAMMPS (Plimpton, 1995), but is implemented within a code that was designed for execution on GPUs from the start (Anderson et al., 2008). The software supports short-ranged pair force and bond force fields and achieves optimal GPU performance using an autotuning algorithm. We are able to demonstrate equivalent or superior scaling on up to 3375 GPUs in Lennard-Jones and dissipative particle dynamics (DPD) simulations of up to 108 million particles. GPUDirect RDMA capabilities in recent GPU generations provide better performance in full double precision calculations. For a representative polymer physics application, HOOMD-blue 1.0 provides an effective GPU vs. CPU node speed-up of 12.5x. (C) 2015 Elsevier B.V. All rights reserved.

引用

页码：97 / 107

页数：11

共 50 条

[1]

Anderson J.A., 2013, ARXIV13085587

[2] General purpose molecular dynamics simulations fully implemented on graphics processing units [J].

Anderson, Joshua A. ;

Lorenz, Chris D. ;

Travesset, A. .

JOURNAL OF COMPUTATIONAL PHYSICS, 2008, 227 (10) :5342-5359

[3]

[Anonymous], THESIS

[4]

Bell Nathan, 2012, GPU computing gems Jade edition, P359, DOI DOI 10.1016/B978-0-12-385963-1.00026-5

[5] Effects of supercoiling on enhancer-promoter contacts [J].

Benedetti, Fabrizio ;

Dorier, Julien ;

Stasiak, Andrzej .

NUCLEIC ACIDS RESEARCH, 2014, 42 (16) :10425-10432

[6] Implementing molecular dynamics on hybrid high performance computers - Particle-particle particle-mesh [J].

Brown, W. Michael ;

Kohlmeyer, Axel ;

Plimpton, Steven J. ;

Tharrington, Arnold N. .

COMPUTER PHYSICS COMMUNICATIONS, 2012, 183 (03) :449-459

[7] Implementing molecular dynamics on hybrid high performance computers - short range forces [J].

Brown, W. Michael ;

Wang, Peng ;

Plimpton, Steven J. ;

Tharrington, Arnold N. .

COMPUTER PHYSICS COMMUNICATIONS, 2011, 182 (04) :898-911

[8]

Carter Edwards H., 2014, J PARALLEL DISTRIB C

[9] Highly accelerated simulations of glassy dynamics using GPUs: Caveats on limited floating-point precision [J].

Colberg, Peter H. ;

Hoefling, Felix .

COMPUTER PHYSICS COMMUNICATIONS, 2011, 182 (05) :1120-1129

[10]

Deng H., 2011, 2011 40 INT C PAR PR, P191

← 1 2 3 4 5 →