A GPU Accelerated Framework for Partitioned Solution of Fluid-Structure Interaction Problems

We present a GPU-accelerated solver for the partitioned solution of fluid-structure interaction (FSI) problems. Independent scalable fluid and structure solvers are coupled by a library which handles the inter-code data communication, mapping and equation coupling. A coupling strategy is incorporated which allows accelerating expensive components of the coupled framework by offloading them to GPUs. To prove the efficiency of the proposed coupling strategy in conjunction with the offloading scheme, we present a numerical performance analysis for a complex test case in the filed of biomedical engineering. The numerical experiments demonstrate an excellent speed-up in the accelerated kernels (up to 133 times) which results in 6 to 8 times faster overall simulations. In addition, we observed a very good reduction in total simulation time by increasing the exploited compute nodes up to 8 (complete machine capacity).We thank the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for supporting this work by funding - EXC2075 – 390740016 under Germany’s Excellence Strategy. We acknowledge the support by the Stuttgart Center for Simulation Science (SimTech). This work was also financially supported by • priority program 1648 - Software for Exascale Computing 214 (ExaFSA - Exascale Simulation of Fluid-Structure-Acoustics Interactions) of the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), • Ministerio de Economía y Competitividad, Secretaría de Estado de Investigacion, Desarrollo e ´ Innovacion, Spain (ENE2017-88697-R). ´ The performance measurements were carried out on the Vulcan cluster at the High-Performance Computing Center Stuttgart (HLRS). The authors wish to thank HLRS for compute time and technical support.Peer ReviewedPostprint (published version