SCEC2023 Plenary Talk, Computational Science (CS)

Application of the EarthQuake SIMulation (EQSIM) framework for GPU-accelerated computers to San Francisco Bay Area earthquake simulations

David B. McCallen, & Arben Pitarka

Oral Presentation

2023 SCEC Annual Meeting, SCEC Contribution #13301
With the continued rapid advancements in high-performance computing hardware, computational ecosystems, and application software workflows, high-fidelity scientific and engineering simulations are undergoing transformational change. The ability to execute three-dimensional simulation models of unprecedented size is removing the historical need to subdivide and execute piecemeal simulations of complex coupled processes. With appropriately developed and tested computational workflows, simulations with hundreds of billions to trillions of grid points and rigorous coupling of the domains of geoscience and engineering are becoming computationally achievable. The EarthQuake SIMulation (EQSIM) multidisciplinary HPC framework for regional-scale, fault-to-structure earthquake simulations has recently been developed under the U.S. Department of Energy Exascale Computing Project. EQSIM was created in preparation for DOE's new GPU-accelerated exaflop platforms with an overall objective of removing computational barriers to high-fidelity regional simulations. In this case study presentation, the computational algorithms and workflow strategies embedded in EQSIM for massively parallel computations are described, and examples of computational performance on broad-band simulations of earthquake ground motions and associated infrastructure response in the San Francisco Bay Area are summarized. Finally, compelling potential early use cases for high-fidelity regional simulations in engineering risk assessments are discussed along with case-specific needs for enhanced geophysical data to appropriately constrain key model parameters.