Momentum significantly accelerates convergence in Federated Learning, especially under data heterogeneity.