Title :
Optimization of OpenFOAM´s linear solvers on emerging multi-core platforms
Author :
Jamshidi, Zahra ; Khunjush, Farshad
Author_Institution :
Sch. of Electr. & Comput. Eng., Shiraz Univ., Shiraz, Iran
Abstract :
In this paper, a popular and open source CFD package, called OpenFOAM, has been ported and optimized on three different multi-core platforms. These platforms include the Cell Broadband Engine, NVIDIA GPU, and an Intel quad-core Xeon CPU. For this, two prevalent linear solvers of this package (i.e., CG and Diagonal solvers) have been selected and optimized on the above-mentioned platforms. Our results show maximum speedups of 15.5 and 345 for the CG and Diagonal solvers, respectively. Moreover, overall results demonstrate that both solvers scale well with various data sizes and thread numbers.
Keywords :
computational fluid dynamics; multiprocessing systems; optimisation; Cell Broadband Engine; Intel quad-core Xeon CPU; NVIDIA GPU; OpenFOAM; linear solvers; multicore platforms; open source CFD package; optimization; Computer architecture; Graphics processing unit; Instruction sets; Libraries; Microprocessors; Optimization; Sparse matrices;
Conference_Titel :
Communications, Computers and Signal Processing (PacRim), 2011 IEEE Pacific Rim Conference on
Conference_Location :
Victoria, BC
Print_ISBN :
978-1-4577-0252-5
Electronic_ISBN :
1555-5798
DOI :
10.1109/PACRIM.2011.6033001
