Providing high-quality “actionable information” for strategic risk analysis is amongst the primary goals of the U.S. Dept. of Energy Exascale Earth System Model (E3SM). The simulation speed required to generate high-quality localized predictions at seasonal-to-decadal time scales is very high. In this talk we highlight some algorithmic design decisions that combine new research with classical numerical methods to enable E3SM’s ultra-high resolution configuration to achieve exascale performance and win the inaugural Gordon Bell Prize for Climate in 2023.

Our design strategies tailor mathematical methods to both the unique features of the application space and to the heterogeneous computing architectures of exascale supercomputers. Ultimately, these efforts doubled the speed of the most computationally demanding component of E3SM, its atmosphere model. We will also discuss new and ongoing research associated with opportunities afforded by these performance gains.