High Performance Matrix Inversion on a Multi-core Platform with Several GPUs