Performance comparison of Lattice Boltzmann fluid flow simulation using OpenCL and CUDA frameworks
University of Novi Sad, Faculty of Sciences, Novi Sad, Serbia
4th International Conference on Information Society and Technology (ICIST 2014), 2014
@article{tekic2014performance,
title={Performance comparison of Lattice Boltzmann fluid flow simulation using OpenCL and CUDA frameworks},
author={Teki{‘c}, Jelena and Teki{‘c}, Predrag and Rackovi{‘c}, Milo{v{s}}},
journal={ICIST 2014}
}
This paper presents performance comparison, of the lid-driven cavity flow simulation, with Lattice Boltzmann method, example, between CUDA and OpenCL parallel programming frameworks. CUDA is parallel programming model developed by NVIDIA for leveraging computing capabilities of their products. OpenCL is an open, royalty free, standard developed by Khronos group for parallel programming of heterogeneous devices (CPU’s, GPU’s, …) from different vendors. OpenCL promises portability of the developed code between heterogeneous devices, but portability has performance penalty. We investigate performance downside of portable OpenCL code comparing to similar CUDA code run on the NVIDIA graphic cards. Lid-driven cavity flow benchmark code, for both examples, has been written in Java programming language, and uses open source libraries to communicate with OpenCL and CUDA. Results of simulations for different grid sizes (from 128 to 896) have been presented and analyzed. Simulations have been carried out on an NVIDIA GeForce GT 220 GPU.
January 2, 2015 by hgpu
