Simulating multi-self-similar spatiotemporal models with CUDA