Using GPUs for Gravity: A potential game changer for numerical relativity

Numerical relativity simulations permit the study of strong gravity regimes around black holes and in the early universe. I will describe the challenges we have faced in upgrading our current numerical relativity code (GRChombo), to a GPU supported version (GRTeclyn), built on top of the AMReX framework. Contrary to our rather pesimistic initial expectations, this effort has paid off - the speed-ups that we have achieved with the new framework are seriously impressive, and should have a game changing impact on the physics that we can study going forward. This puts us in strong position for interpreting the non-linear regime of gravity using data from current and planned gravitational wave detectors.