SCEC Award Number 12142 View PDF
Proposal Category Individual Proposal (Data Gathering and Products)
Proposal Title Accelerating CyberShake SGT Calculations on Heterogeneous Supercomputers
Investigator(s)
Name Organization
Yifeng Cui University of California, San Diego
Other Participants Zhou, Jun, GSR at UCSD
SCEC Priorities 6e, 4e, 6a SCEC Groups CME, GMP, CS
Report Due Date 03/15/2013 Date Report Submitted N/A
Project Abstract
 With SCEC 2012 core support, we have developed a highly scalable and efficient GPU-based finite-difference code (AWP-ODC-GPU, hereafter abbreviated AWP) for earthquake simulation through high throughput, memory locality, communication reduction and communication / computation overlap. We achieved perfect linear speedup on Cray XK7 Titan at ORNL and NCSA’s Blue Waters system. AWP’s excellent performance is demonstrated by simulating realistic 0-10 Hz earthquake ground motions, 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.
Intellectual Merit The successful development of this GPU-based code provides realistic ground motion estimates from large surface rupturing earthquakes, that provide a critical basis for developing improved Ground Motion Prediction Equations (GMPEs) for large crustal earthquakes at short distances from the fault, where data is sparse or absent. The physics-based calculation of California-wide hazard curves up to 1Hz provides the basis for yet another check of the GMPEs and understanding their limitations. Deterministic earthquake scenario simulation of wave propagation at frequencies up to 10 Hz with low near-surface wave speeds is important for earthquake engineers to scrutinize for potential use in design, to illuminate crustal scattering processes and to provide a reality check for the deterministic-stochastic methods on the SCEC Broadband Platform.
Broader Impacts Improved seismic modeling will help inform urban planning, seismic design provisions in building codes, performance-based earthquake engineering, and disaster planning. The use of the AWP-SGT code is expected to save up to 500 million hours of computation required for the proposed statewide CyberShake 3.0 model, in addition to reducing dramatically the time-to-solution. This effort will expand our outreach plan to take advantage of the SCEC implementation interface to develop a dialog among user communities, and eventually further Earth system research and provide improved seismic hazard analysis and a road map to engineer safer structures. Finally, graduate students will gain valuable research collaboration experience from this effort.
Exemplary Figure Figure 7: Snapshots of 10-Hz rupture propagation (slip rate) and surface wavefield (strike-parallel component) for a crustal model (top) without and (bottom) with a statistical model of small-scale heterogeneities. The displayed geometrical complexities on the fault were included in the rupture simulation. The associated synthetic strike-parallel component seismograms are superimposed as black traces on the surface at selected sites. The part of the crustal model located in front of the fault has been lowered for a better view. Note the strongly scattered wavefield in the bottom snapshot due to the small-scale heterogeneities. (Credits: Simulation: Efecan Poyraz and Yifeng Cui of SDSC, Kyle Withers and Kim Olsen of SDSU. Visualization: Amit Chourasia of SDSC).