GPU-based scale-resolving simulations of the chaotic cardiovascular flows*

Blood flow in the heart and large vessels presents a complex vortical character which involves a rich cascade of spatial and temporal scales. For instance, ejection of oxygenated blood out of the left ventricle results in a relatively high Reynolds number flow (Re=5000-10000) which is prone to a turbulence signature through interaction with the complex aortic geometry. Besides vessel curvature and subtle surface features, turbulence in the aorta may further be enhanced by a disease scenario e.g. atherosclerosis, and/or a present medical device. Scale-resolving computational fluid dynamics (CFD) simulations can be used to provide a detailed view of this turbulent flow. We present a GPU-based incompressible Navier-Stokes solver for scale-resolving simulations of blood flow in patient-specific aortic geometries. The flow solver leverages an optimised MPI-CUDA implementation and dedicated numerics, to coup with grid sizes beyond one billion. The geometric features are extracted from computed tomography (CT) angiograms and are integrated into the solver using a sharp-interface immersed boundary method. Using the present GPU-based solver, we show that intensive simulations matching O(100M) voxels which often require O(10K) CPU cores can now be performed with only O(10) P100 GPUs.