Performance Improvement of TOUGH2 Simulation with Graphics Processing Unit
Faculty of Engineering, Kyushu University, Nishi-ku, Fukuoka, JAPAN 819-0395
TOUGH Symposium, 2012
@article{shimotoku2012performance,
title={Performance Improvement of TOUGH2 Simulation with Graphics Processing Unit},
author={Shimotoku, Yusuke and Tanaka, Toshiaki and Itoi, Ryuichi},
year={2012}
}
We tried to accelerate the computational speed of TOUGH2 simulation by introducing a linear computation routine using a Graphics Processing Unit (GPU). Libraries for GPU computation were introduced, and new solvers for linear equations were developed. Out of those, CLLUSTB, an ILU preconditioned BiCGSTAB solver made with the CULA Sparse, demonstrated good performance both in speed and accuracy. We introduced the new solvers into TOUGH2 to improve its performance and, for a sample problem, conducted a natural-state three-dimensional simulation of a geothermal reservoir. As a result, the routine with GPU was 1.44 times faster in matrix calculation, and 1.34 times faster in simulation, than the computed results with the iterative solver DLUSTB, originally built into TOUGH2.
September 10, 2012 by hgpu
