We present a scalable solver for density driven flow in porous media
based on operator splitting and a Discontinuous Galerkin discretization using
the frameworks DUNE and PDELab. In particular, we focus on the efficient
exploitation of current and next generation HPC architectures. For this purpose,
we extend the existing parallel computation support in DUNE (based
on classical MPI domain partitioning) to a hybrid model with shared memory
parallelism on multi-threaded CPU nodes and accelerators like GPGPUs or
Intel MIC. Moreover, we investigate the effects of leveraging the SIMD capabilities
of current hardware at the different solver stages (assembly / linear
algebra).