ECCOMAS 2024

Performance Portable Design of an Exascale Multiphysics Simulator

  • Settgast, Randolph (Lawrence Livermore National Laboratory)
  • Tobin, William (Lawrence Livermore National Laboratory)
  • Magri, Victor (Lawrence Livermore National Laboratory)
  • Castelletto, Nicola (Lawrence Livermore National Laboratory)

Please login to view abstract download link

Carbon capture and storage (CCUS) is a key component in global efforts to reduce carbon emissions. The evaluation of CCUS candidate sites requires predictive simulation capabilities involving coupled poromechanics, multicomponent flow, with thermal effects. In this context, we present an overview of the portability strategy of the GEOS simulation framework[1]. The strategies for solving linear systems of equations resulting from coupled physics will be presented for both matrix-based and matrix-free methods. We will also provide additional details on constructing performant kernels for use in matrix-free methods. Performance results of validation problems will be presented with a specific focus on GPU performance and weak scaling.