Accelerating tracer transport in FESOM-2 with GPU's

Within the ESiWACE-2 service on porting applications to pre-exascale machines, we have collaborated with the FESOM-2 team to accelerate parts of their model on GPU's. FESOM-2 is a finite-volume ocean and sea-ice model based on unstructured meshes. We have used the OpenACC programming model to speed up the transport of tracers in FESOM-2. In this talk, I will share the results of this effort, describe our collaboration with the FESOM team and present some technical aspects of the implementation. Finally, I will iterate on the lessons learned and the road ahead for FESOM-2 towards exascale.