DualSPHysics, new GPU computing on SPH models

Benedict D. Rogers; T Rung; C Ulrich

DualSPHysics, new GPU computing on SPH models

Benedict D. Rogers, T Rung (Editor), C Ulrich (Editor)

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

Abstract

The capacity of a single processor is insufficient torun huge 3D simulations with Smoothed ParticleHydrodynamics (SPH) methods. Here we present thedevelopment of a new code named DualSPHysics that can beexecuted either on either Central Processing Units (CPU) oron Graphics Processor Units (GPUs). Thus, the parallelisationapproach using the Compute Unified Device Architecture(CUDA) of nVidia is described. Simulations with more thanone million particles on a single GPU card exhibit speedups ofup to two orders of magnitude over using a single-core CPU.The DualSPHysics code is validated with the experimentaldata corresponding to a dam break flow impacting on anobstacle and the agreement between the numerical andexperimental results is analysed. This new technology makespossible the study of real-life engineering problems at areasonable computational cost such as fluid-structureinteraction.

Original language	English
Title of host publication	Proc. 6th International SPHERIC Workshop
Editors	T Rung, C Ulrich
Pages	348-354
Number of pages	7
Publication status	Published - 2011
Event	6th International SPHERIC Workshop - TUHH Hamburg Duration: 8 Jun 2011 → 10 Jun 2011

Conference

Conference	6th International SPHERIC Workshop
City	TUHH Hamburg
Period	8/06/11 → 10/06/11

Cite this

@inproceedings{6e9a399d773940bda2b275e90a0bb27d,

title = "DualSPHysics, new GPU computing on SPH models",

abstract = "The capacity of a single processor is insufficient torun huge 3D simulations with Smoothed ParticleHydrodynamics (SPH) methods. Here we present thedevelopment of a new code named DualSPHysics that can beexecuted either on either Central Processing Units (CPU) oron Graphics Processor Units (GPUs). Thus, the parallelisationapproach using the Compute Unified Device Architecture(CUDA) of nVidia is described. Simulations with more thanone million particles on a single GPU card exhibit speedups ofup to two orders of magnitude over using a single-core CPU.The DualSPHysics code is validated with the experimentaldata corresponding to a dam break flow impacting on anobstacle and the agreement between the numerical andexperimental results is analysed. This new technology makespossible the study of real-life engineering problems at areasonable computational cost such as fluid-structureinteraction.",

author = "Rogers, {Benedict D.} and T Rung and C Ulrich",

year = "2011",

language = "English",

pages = "348--354",

editor = "T Rung and C Ulrich",

booktitle = "Proc. 6th International SPHERIC Workshop",

note = "6th International SPHERIC Workshop ; Conference date: 08-06-2011 Through 10-06-2011",

}

TY - GEN

T1 - DualSPHysics, new GPU computing on SPH models

AU - Rogers, Benedict D.

A2 - Rung, T

A2 - Ulrich, C

A2 - Rung, T

A2 - Ulrich, C

PY - 2011

Y1 - 2011

N2 - The capacity of a single processor is insufficient torun huge 3D simulations with Smoothed ParticleHydrodynamics (SPH) methods. Here we present thedevelopment of a new code named DualSPHysics that can beexecuted either on either Central Processing Units (CPU) oron Graphics Processor Units (GPUs). Thus, the parallelisationapproach using the Compute Unified Device Architecture(CUDA) of nVidia is described. Simulations with more thanone million particles on a single GPU card exhibit speedups ofup to two orders of magnitude over using a single-core CPU.The DualSPHysics code is validated with the experimentaldata corresponding to a dam break flow impacting on anobstacle and the agreement between the numerical andexperimental results is analysed. This new technology makespossible the study of real-life engineering problems at areasonable computational cost such as fluid-structureinteraction.

AB - The capacity of a single processor is insufficient torun huge 3D simulations with Smoothed ParticleHydrodynamics (SPH) methods. Here we present thedevelopment of a new code named DualSPHysics that can beexecuted either on either Central Processing Units (CPU) oron Graphics Processor Units (GPUs). Thus, the parallelisationapproach using the Compute Unified Device Architecture(CUDA) of nVidia is described. Simulations with more thanone million particles on a single GPU card exhibit speedups ofup to two orders of magnitude over using a single-core CPU.The DualSPHysics code is validated with the experimentaldata corresponding to a dam break flow impacting on anobstacle and the agreement between the numerical andexperimental results is analysed. This new technology makespossible the study of real-life engineering problems at areasonable computational cost such as fluid-structureinteraction.

M3 - Conference contribution

SP - 348

EP - 354

BT - Proc. 6th International SPHERIC Workshop

T2 - 6th International SPHERIC Workshop

Y2 - 8 June 2011 through 10 June 2011

ER -

DualSPHysics, new GPU computing on SPH models

Abstract

Conference

Fingerprint

Cite this