FastIce.jl: A Massively Parallel Ice Flow Model Running on GPUs

Explore a groundbreaking 33-minute conference talk introducing FastIce.jl, a cutting-edge ice flow model designed for massively parallel architectures. Discover how this Julia-based software leverages GPUs and distributed computing to efficiently model complex ice sheet dynamics, including thermo-mechanical interactions. Learn about the modular architecture of FastIce.jl, its core library of components, and monolithic solvers implementing ice flow models. Gain insights into the use of task-based programming in Julia for optimizing GPU kernel execution, CPU functions, and MPI communication. Examine performance testing results from single-node and distributed scaling benchmarks conducted on LUMI, Europe's largest supercomputer, and understand how FastIce.jl's fully differentiable core enables data assimilation pipelines using adjoint sensitivities and automatic differentiation.

FastIce.jl: a massively parallel ice flow model running on GPUs | Räss, Omlin, Werder, Utkin