The presentation investigates various aspects of the implementation of algebraic multigrid methods on GPU-accelerated hybrid architectures. Furthermore, different parallelization strategies are investigated to achieve optimal performance on a broad range of hardware configurations.