Exascale target (capability/capacity)
The currently largest setups already require petascale performance, despite their limited size of the simulated domain (local tsunami in the Palu Bay).
Larger simulated domains and higher resolution will require a further increase in the degrees of freedom.
Higher resolution is particularly required for resolving the acoustic layer in the ocean, which promises new avenues for tsunami early warning.
Extension towards parameter studies and uncertainty quantification will lead to further setups that require exascale performance.
Main benefits expected from large increase in compute resources
Higher resolution to capture a wider range of frequencies for seismic and acoustic waves.
Possibility to tackle domains with larger extent.
Parameter studies and uncertainty quantification are only possible if time to solution for individual runs is reduced by 2-3 orders of magnitude.
Main challenges in harnessing large compute resources
Full support of the elastic-acoustic model (esp. incl. dynamic rupture and boundary conditions) also in the GPU version of SeisSol, ensuring performance portability on upcoming exascale architectures.
Specific actions towards exascale at present time
Porting of SeisSol (incl. elastic-acoustic coupling with necessary boundary conditions and dynamic rupture) to GPU supercomputers.
Extension of the underlying code generator YATeTo towards various heterogeneous architectures.
Optimisation of SeisSol’s local time stepping algorithm.