Limits...
GPUs, a new tool of acceleration in CFD: efficiency and reliability on smoothed particle hydrodynamics methods.

Crespo AC, Dominguez JM, Barreiro A, Gómez-Gesteira M, Rogers BD - PLoS ONE (2011)

Bottom Line: It is demonstrated that the code achieves different speedups with different CUDA-enabled GPUs.The numerical behaviour of the SPH code is validated with a standard benchmark test case of dam break flow impacting on an obstacle where good agreement with the experimental results is observed.Both the achieved speed-ups and the quantitative agreement with experiments suggest that CUDA-based GPU programming can be used in SPH methods with efficiency and reliability.

View Article: PubMed Central - PubMed

Affiliation: EPHYSLAB Environmental Physics Laboratory, Universidade de Vigo, Ourense, Spain. alexbexe@uvigo.es

ABSTRACT
Smoothed Particle Hydrodynamics (SPH) is a numerical method commonly used in Computational Fluid Dynamics (CFD) to simulate complex free-surface flows. Simulations with this mesh-free particle method far exceed the capacity of a single processor. In this paper, as part of a dual-functioning code for either central processing units (CPUs) or Graphics Processor Units (GPUs), a parallelisation using GPUs is presented. The GPU parallelisation technique uses the Compute Unified Device Architecture (CUDA) of nVidia devices. Simulations with more than one million particles on a single GPU card exhibit speedups of up to two orders of magnitude over using a single-core CPU. It is demonstrated that the code achieves different speedups with different CUDA-enabled GPUs. The numerical behaviour of the SPH code is validated with a standard benchmark test case of dam break flow impacting on an obstacle where good agreement with the experimental results is observed. Both the achieved speed-ups and the quantitative agreement with experiments suggest that CUDA-based GPU programming can be used in SPH methods with efficiency and reliability.

Show MeSH
Computational runtime distribution on GPU (Tesla M1060).
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3113801&req=5

pone-0020685-g010: Computational runtime distribution on GPU (Tesla M1060).

Mentions: Figure 10 shows the new time distribution once the NL, PI and SU are performed entirely on the GPU device. Particle interaction times range from 81% (low resolution) to 92% (high resolution) of the total runtime. The percentage of computational time used for NL and SU is larger than observed for CPU calculations, although it decreases when increasing the resolution.


GPUs, a new tool of acceleration in CFD: efficiency and reliability on smoothed particle hydrodynamics methods.

Crespo AC, Dominguez JM, Barreiro A, Gómez-Gesteira M, Rogers BD - PLoS ONE (2011)

Computational runtime distribution on GPU (Tesla M1060).
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3113801&req=5

pone-0020685-g010: Computational runtime distribution on GPU (Tesla M1060).
Mentions: Figure 10 shows the new time distribution once the NL, PI and SU are performed entirely on the GPU device. Particle interaction times range from 81% (low resolution) to 92% (high resolution) of the total runtime. The percentage of computational time used for NL and SU is larger than observed for CPU calculations, although it decreases when increasing the resolution.

Bottom Line: It is demonstrated that the code achieves different speedups with different CUDA-enabled GPUs.The numerical behaviour of the SPH code is validated with a standard benchmark test case of dam break flow impacting on an obstacle where good agreement with the experimental results is observed.Both the achieved speed-ups and the quantitative agreement with experiments suggest that CUDA-based GPU programming can be used in SPH methods with efficiency and reliability.

View Article: PubMed Central - PubMed

Affiliation: EPHYSLAB Environmental Physics Laboratory, Universidade de Vigo, Ourense, Spain. alexbexe@uvigo.es

ABSTRACT
Smoothed Particle Hydrodynamics (SPH) is a numerical method commonly used in Computational Fluid Dynamics (CFD) to simulate complex free-surface flows. Simulations with this mesh-free particle method far exceed the capacity of a single processor. In this paper, as part of a dual-functioning code for either central processing units (CPUs) or Graphics Processor Units (GPUs), a parallelisation using GPUs is presented. The GPU parallelisation technique uses the Compute Unified Device Architecture (CUDA) of nVidia devices. Simulations with more than one million particles on a single GPU card exhibit speedups of up to two orders of magnitude over using a single-core CPU. It is demonstrated that the code achieves different speedups with different CUDA-enabled GPUs. The numerical behaviour of the SPH code is validated with a standard benchmark test case of dam break flow impacting on an obstacle where good agreement with the experimental results is observed. Both the achieved speed-ups and the quantitative agreement with experiments suggest that CUDA-based GPU programming can be used in SPH methods with efficiency and reliability.

Show MeSH