Summaries - Office of Research & Innovation
Back Utilizing Graphics Processing Units for Dynamic Rupture Earthquake Simulations
|Division||Graduate School of Engineering & Applied Science|
|Investigator(s)||Kozdon, Jeremy E.|
|Sponsor||National Science Foundation (NSF)|
The instrumented earthquake era is relatively short compared with the return interval of large, damaging earthquakes. Thus, modeling efforts are required to assess the probability of different earthquake scenarios and their associated hazard. Of particular importance is the effect of geometric complexity on hazard, such as the probability that a rupture will take one branch of a fault over another or how likely it is for a rupture to jump to a neighboring fault.
These questions and many others can only be answered using fully coupled, physics-based, and self-consistent earthquake rupture dynamics simulations. This requires efficient, accurate, and stable numerical methods. Though there are several parallel codes that implement rupture dynamics, to date none of the methods can be used with newly available accelerator technologies (e.g., graphics processing units, GPUs, and Intel Xeon Phi coprocessors). This is particularly important given that three of the five NSF XSEDE resources as well as about $10\%$ of the Top500 supercomputers have accelerators. Thus, applications that do not consider these technologies will have sub-optimal performance in the near future. Under this proposal a new dynamic rupture method will be developed for the GPU-based machines. The foundation for this work is the high-order accurate finite difference method for rupture dynamics the PI developed under the CI TraCS fellowship. In addition to being the first time that rupture dynamics has been performed on the GPU, the numerical method will also allow for the use of mesh refinement to capture the multiscale nature of earthquake rupture. It was shown in the PI’s CI TraCS work that dynamic mesh refinement can drastically reduce the computational demands of earthquake rupture simulations. For all these tasks, the main computational concerns are boundaries and interfaces, which require special handling on the GPU in order to ensure efficient GPU memory management. As a starting tenure-track faculty member, the equipment procured under this proposal will support the PI in developing a robust and relevant research program with the latest computing technologies.
|Publications||Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal|
|Data||Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal|