07-12, 14:22–14:44 (Europe/Amsterdam), Function (4.1)
CliMA's Earth System Model (ESM) is designed to automatically learn from diverse data sources, such as global observations and targeted high-resolution simulations, to reduce and quantify uncertainties in climate predictions. The components of the ESM software stack include an atmosphere model, a radiation model, a land model, a coupler, among others.
CliMA's Earth System Model (ESM) is designed to automatically learn from diverse data sources, such as global observations and targeted high-resolution simulations, to reduce and quantify uncertainties in climate predictions. The components of the ESM software stack include an atmosphere model, a radiation model, a land model, a coupler, among others. ClimaComms.jl provides the high-level communications interface and computational backend for various components of the ESM software stack.
This talk describes some of the most expensive communication patterns encountered in the atmospheric solver (Direct Stiffness Summation), and how the communication interface is implemented to provide support for various computational backends (single-threaded/multi-threaded CPUs, GPUs) and transports (Serial, MPI), thereby enabling the user to run distributed simulations on a wide range of architectures.
