Towards a Distributed GPU-Accelerated Matrix Inversion