Large-Scale Thermo-Mechanical Ice Flow Modelling Using Julia on GPU
DescriptionThe current climate change triggers a critical acceleration of ice loss at global scale, may it happen in Antarctica and Greenland or in mountain regions such as the Alps. This situation ultimately poses a threat for coastal regions affected by sea-level rise or in valley-glaciers subject to surges and collapses of ice masses. Numerical modelling permits to understand and predict the evolution of these complex natural systems resolving the underlying physical processes in three dimensions. Moreover, high spatial and temporal resolution is crucial to capture rapid changes in the system leading to the formation of, e.g., ice streams or collapse features. In this work, we present innovative software tools which provide a way forward in ice dynamics and computational Earth sciences by exploiting massively parallel supercomputing on graphics processing units (GPUs) and scalable iterative solvers. We use the Julia language because it features high-level capabilities while facilitating high performance and portability amongst multiple backends (e.g., multi-core CPUs, and NVIDIA and AMD GPUs). We assess the performance of the GPU-accelerated implementation of the ice flow model, demonstrate results of benchmarks, and share our experiences using Julia for HPC.
TimeWednesday, June 2814:00 - 14:30 CEST
Computer Science, Machine Learning, and Applied Mathematics