Multi-level Parallelism for Incompressible Flow Computations on GPU Clusters