Informatics
Karlsruher Institut für Technologie (KIT)
Steinbuch Centre for Computing
Using the PI’s linear algebra library Ginkgo, parts of OpenFOAM were successfully ported to run on the GPU. This project intends to demonstrate larger speed-ups on a large-scale application by porting more parts including non-linear portions to the GPU, and by tuning the application’s setup for GPUs. The latter is especially important since the solver type and its runtime controls like timestep size, outer and inner iterations are typically optimized for CPU performance. We intend to modify the application’s setup to be GPU-friendly while keeping the accuracy of the CPU setup. The work is a joint initiative of the PI and Upstream CFD GmbH in the framework of the EXASIM project. The application is the Grand Challenge that Upstream CFD contributes as partner and application specialist to the exaFOAM project. It is the WMLES of an aircraft in landing configuration as defined by the AIAA High-Lift Prediction Workshop 2023.