Physics-based Seismic Hazard Analysis on Petascale Heterogeneous Supercomputers

Yifeng Cui, Efecan Poyraz, Kim B. Olsen, Jun Zhou, Kyle B. Withers, Scott Callaghan, Jeff Larkin, Clark Guest, Dong Ju Choi, Amit Chourasia, Zheqiang Shi, Steven M. Day, Philip J. Maechling, & Thomas H. Jordan

Published November 2013, SCEC Contribution #1753

We have developed a highly scalable and efficient GPU-based finite-difference code (AWP) for earthquake simulation through high throughput, memory locality, communication reduction and communication / computation overlap, achieving perfect linear speedup on Cray XK7 Titan at ORNL and NCSA’sBlue Waters system. AWP’s excellent performance is demonstrated by simulating realistic 0-10 Hz earthquake ground motions, as required by building engineering design, through small-scale complexity in the fault surface and surrounding crustal structure. Moreover, we show that AWP provides a speedup in key strain tensor calculations critical to probabilistic seismic hazard analysis by a factor of 110. This achievement, coupled with improved co- scheduling capabilities of our workflow-managed systems, makes a statewide hazard model a goal reachable with existing supercomputers. The performance of the GPU-based AWP is expected to take physics-based seismic hazard analysis to a new level using Petascale heterogeneous computing resources, saving millions of core-hours over the next few years.

Cui, Y., Poyraz, E., Olsen, K. B., Zhou, J., Withers, K. B., Callaghan, S., Larkin, J., Guest, C., Choi, D., Chourasia, A., Shi, Z., Day, S. M., Maechling, P. J., & Jordan, T. H. (2013, 11). Physics-based Seismic Hazard Analysis on Petascale Heterogeneous Supercomputers. Oral Presentation at SC13. doi: 10.1145/2503210.2503300.