Technical report detail

Evaluating realistic tsunami simulations with SWE model on GPU-enabled clusters

by Juraj Kardos

Tsunamis are devastating natural disasters, potentially able to create a lot of damage. In order to predict such events and their behavior fast and detailed simulations of tsunamis are required. In this paper we evaluate the ability of SWE shallow water numerical model to perform efficient and realistic tsunami simulations. We used SWE model to reproduce 2004 Great Sumatra-Andaman and 2011 Tohoku tsunami. Nowadays, many of the Top500 supercomputers are using GPU accelerators to enhance the peak performance. The GPU-enabled clusters Tesla-CMC and Todi Cray XK-7 were benchmarked using different number of processes and GPUs to obtain the most efficient and the most fast SWE configurations. Using all 3 available GPUs of Tesla-CMC, we observed the maximum speedup of 158x in comparison to single-core CPU version. The most efficient configuration was the one with a single GPU, achieving speedup of 58x. Performance results for "Todi" Cray XK-7 are also presented for larger number of compute nodes.

Technical report 2013/03, July 2013

BibTex entry

@techreport{13evaluating, author = {Juraj Kardos}, title = {Evaluating realistic tsunami simulations with SWE model on GPU-enabled clusters}, institution = {University of Lugano}, number = {2013/03}, year = 2013, month = jul }