Multi-GPU and Multi-CPU Parallelization for Interactive Physics Simulations

Everton Hermann; Bruno Raffin; François Faure; Thierry Gautier; Jérémie Allard

inria-00502448, version 1

Multi-GPU and Multi-CPU Parallelization for Interactive Physics Simulations

Everton Hermann^b^, 1, 2, Bruno Raffin ², François Faure ()^a^, 1, 3, Thierry Gautier ()^b^, 2, Jérémie Allard () ⁴

Europar 2010 (2010)

Abstract: Today, it is possible to associate multiple CPUs and multiple GPUs in a single shared memory architecture. Using these resources efficiently in a seamless way is a challenging issue. In this paper, we propose a parallelization scheme for dynamically balancing work load between multiple CPUs and GPUs. Most tasks have a CPU and GPU implementation, so they can be executed on any processing unit. We rely on a two level scheduling associating a traditional task graph partitioning and a work stealing guided by processor affinity and heterogeneity. These criteria are intended to limit inefficient task migrations between GPUs, the cost of memory transfers being high, and to favor mapping small tasks on CPUs and large ones on GPUs to take advantage of heterogeneity. This scheme has been implemented to support the SOFA physics simulation engine. Experiments show that we can reach speedups of 22 with 4 GPUs and 29 with 4 CPU cores and 4 GPUs. CPUs unload GPUs from small tasks making these GPUs more efficient, leading to a "cooperative speedup" greater than the sum of the speedups separatly obtained on 4 GPUs and 4 CPUs.

a – Université Joseph Fourier - Grenoble I
b – INRIA
1: EVASION (INRIA Grenoble Rhône-Alpes / LJK Laboratoire Jean Kuntzmann)
CNRS : UMR5224 – INRIA – Laboratoire Jean Kuntzmann – Institut National Polytechnique de Grenoble (INPG) – Université Joseph Fourier - Grenoble I – Université Pierre-Mendès-France - Grenoble II
2: MOAIS (INRIA Grenoble Rhône-Alpes / LIG Laboratoire d'Informatique de Grenoble)
INRIA – Université Joseph Fourier - Grenoble I – Institut polytechnique de Grenoble (Grenoble INP) – Université Pierre-Mendès-France - Grenoble II – CNRS : UMR5217
3: Laboratoire Jean Kuntzmann (LJK)
CNRS : UMR5224 – Université Joseph Fourier - Grenoble I – Université Pierre-Mendès-France - Grenoble II – Institut Polytechnique de Grenoble - Grenoble Institute of Technology
4: ALCOVE (INRIA Lille - Nord Europe)
INRIA – CNRS : UMR8022 – Université Lille I - Sciences et technologies – IRCICA

inria-00502448, version 1
http://hal.inria.fr/inria-00502448
oai:hal.inria.fr:inria-00502448
From: François Faure <>
Submitted on: Thursday, 15 July 2010 01:07:30
Updated on: Monday, 17 September 2012 15:40:17

See detailed view

Associated documents

PDF :

Export

Bibtex EndNote TEI RefWorks

%% inria-00502448, version 1
%% http://hal.inria.fr/inria-00502448
@inproceedings{hermann:inria-00502448,
    hal_id = {inria-00502448},
    url = {http://hal.inria.fr/inria-00502448},
    title = {{Multi-GPU and Multi-CPU Parallelization for Interactive Physics Simulations}},
    author = {Hermann, Everton and Raffin, Bruno and Faure, Fran{\c c}ois and Gautier, Thierry and Allard, J{\'e}r{\'e}mie},
    abstract = {{Today, it is possible to associate multiple CPUs and multiple GPUs in a single shared memory architecture. Using these resources efficiently in a seamless way is a challenging issue. In this paper, we propose a parallelization scheme for dynamically balancing work load between multiple CPUs and GPUs. Most tasks have a CPU and GPU implementation, so they can be executed on any processing unit. We rely on a two level scheduling associating a traditional task graph partitioning and a work stealing guided by processor affinity and heterogeneity. These criteria are intended to limit inefficient task migrations between GPUs, the cost of memory transfers being high, and to favor mapping small tasks on CPUs and large ones on GPUs to take advantage of heterogeneity. This scheme has been implemented to support the SOFA physics simulation engine. Experiments show that we can reach speedups of 22 with 4 GPUs and 29 with 4 CPU cores and 4 GPUs. CPUs unload GPUs from small tasks making these GPUs more efficient, leading to a "cooperative speedup" greater than the sum of the speedups separatly obtained on 4 GPUs and 4 CPUs.}},
    language = {English},
    affiliation = {EVASION - INRIA Grenoble Rh{\^o}ne-Alpes / LJK Laboratoire Jean Kuntzmann , MOAIS - INRIA Grenoble Rh{\^o}ne-Alpes / LIG Laboratoire d'Informatique de Grenoble , Laboratoire Jean Kuntzmann - LJK , ALCOVE - INRIA Lille - Nord Europe},
    booktitle = {{Europar 2010}},
    address = {Ischia-Naples, Italy},
    audience = {international },
    year = {2010},
    month = Sep,
    pdf = {http://hal.inria.fr/inria-00502448/PDF/europar10.pdf},
}

<?xml version="1.0" encoding="UTF-8"?>
<listBibl xmlns="http://www.tei-c.org/ns/1.0" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML">
<listBibl type="inproceedings"><head>Peer-reviewed conferences/proceedings</head>
<biblStruct type="inproceedings" xml:id="inria-00502448"><analytic>
<author><persName><forename>Everton</forename><surname>Hermann</surname></persName>
<affiliation>
<orgName type="laboratory">INRIA Grenoble Rhône-Alpes / LJK Laboratoire Jean Kuntzmann</orgName>
<orgName type="team">EVASION</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR5224</orgName>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">Laboratoire Jean Kuntzmann</orgName>
<orgName type="institution">Institut National Polytechnique de Grenoble (INPG)</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
</affiliation>
<affiliation>
<orgName type="laboratory">INRIA Grenoble Rhône-Alpes / LIG Laboratoire d'Informatique de Grenoble</orgName>
<orgName type="team">MOAIS</orgName>
<country>FR</country>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Institut polytechnique de Grenoble (Grenoble INP)</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
<orgName type="institution">CNRS : UMR5217</orgName>
</affiliation>
</author>
<author><persName><forename>Bruno</forename><surname>Raffin</surname></persName>
<affiliation>
<orgName type="laboratory">INRIA Grenoble Rhône-Alpes / LIG Laboratoire d'Informatique de Grenoble</orgName>
<orgName type="team">MOAIS</orgName>
<country>FR</country>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Institut polytechnique de Grenoble (Grenoble INP)</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
<orgName type="institution">CNRS : UMR5217</orgName>
</affiliation>
</author>
<author><persName><forename>François</forename><surname>Faure</surname></persName>
<affiliation>
<orgName type="laboratory">INRIA Grenoble Rhône-Alpes / LJK Laboratoire Jean Kuntzmann</orgName>
<orgName type="team">EVASION</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR5224</orgName>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">Laboratoire Jean Kuntzmann</orgName>
<orgName type="institution">Institut National Polytechnique de Grenoble (INPG)</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
</affiliation>
<affiliation>
<orgName type="laboratory">LJK</orgName>
<orgName type="team">Laboratoire Jean Kuntzmann</orgName>
<country>FR</country>
<orgName type="institution">CNRS : UMR5224</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
<orgName type="institution">Institut Polytechnique de Grenoble - Grenoble Institute of Technology</orgName>
</affiliation>
</author>
<author><persName><forename>Thierry</forename><surname>Gautier</surname></persName>
<affiliation>
<orgName type="laboratory">INRIA Grenoble Rhône-Alpes / LIG Laboratoire d'Informatique de Grenoble</orgName>
<orgName type="team">MOAIS</orgName>
<country>FR</country>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">Université Joseph Fourier - Grenoble I</orgName>
<orgName type="institution">Institut polytechnique de Grenoble (Grenoble INP)</orgName>
<orgName type="institution">Université Pierre-Mendès-France - Grenoble II</orgName>
<orgName type="institution">CNRS : UMR5217</orgName>
</affiliation>
</author>
<author><persName><forename>Jérémie</forename><surname>Allard</surname></persName>
<affiliation>
<orgName type="laboratory">INRIA Lille - Nord Europe</orgName>
<orgName type="team">ALCOVE</orgName>
<country>FR</country>
<orgName type="institution">INRIA</orgName>
<orgName type="institution">CNRS : UMR8022</orgName>
<orgName type="institution">Université Lille I - Sciences et technologies</orgName>
<orgName type="institution">IRCICA</orgName>
</affiliation>
</author>
<title type="main" level="a">Multi-GPU and Multi-CPU Parallelization for Interactive Physics Simulations</title></analytic><monogr><meeting>Europar 2010</meeting><imprint><pubPlace><settlement>Ischia-Naples</settlement><country>Italy</country></pubPlace><biblScope type="publicationDate">2010-09-00</biblScope></imprint></monogr><idno type="HALid">http://hal.inria.fr/inria-00502448</idno><idno type="HALFile">http://hal.inria.fr/inria-00502448/PDF/europar10.pdf</idno></biblStruct>
</listBibl>
</listBibl>