CISL/CGD Seminar: Earth System Modeling in the Exascale Era: The E3SM Atmosphere Model in C++/Kokkos (EAMxx)
Mesa Lab MSR & Virtual
1:00 – 2:00 pm MDT
Abstract
The era of exascale computing has become a reality, driven by the adoption of mixed CPU/GPU architectures that deliver unprecedented performance and efficiency. To harness the power of these advanced systems, the Earth System Modeling community has begun transitioning to GPU-ready models. However, this shift presents significant challenges due to the complexity and scale of existing models. Compounding the difficulty, many Earth System models are written in Fortran, a language that poses considerable obstacles for GPU adaptation.
The DOE Energy Exascale Earth System Model (E3SM) turned this challenge into an opportunity by completely rewriting its atmosphere component. Building on the successful port of the HOMME dynamics core to C++/Kokkos (HOMMExx), the E3SM Atmosphere Model (EAMxx) has been redesigned from the ground up specifically to leverage GPU performance. This talk will explore the successes and challenges faced by the EAMxx team during the transition to C++/Kokkos. It will also highlight key lessons learned and outline the path forward for Earth System models in the exascale era.
Name
Aaron Donahue
Aaron Donahue joined Lawrence Livermore National Laboratory (LLNL) in 2016 after completing his Ph.D. at the University of Notre Dame with a degree in Civil Engineering. His research has focused on the development and analysis of global atmosphere models. His most recent work has been in leading the development of the EAMxx atmosphere driver, a complete rewrite in Kokkos/C++ of the infrastructure that drives the E3SM Atmosphere Model (EAM).