<p>Transient forced vibration analysis of rigid-flexible multibody systems remains one of the core research problems in mechanical system dynamics. For linear systems under assumptions of minor vibrations, minor deformations, and linear elastic materials, the modal superposition method is employed for transient response computation. Finite element discretization is integrated with the Craig-Bampton substructure reduction technique to reduce the degree of freedom of flexible bodies. Data dependencies inherent in solving the system’s generalized coordinate equations during the computation process severely constrain the parallel computational efficiency of the conventional approach. This paper proposes a GPU-accelerated parallel algorithm based on modal superposition to address this limitation. The temporal integration process in the generalized coordinate equation-solving procedure is restructured into a hierarchical parallel accumulation by introducing a parallel prefix sum algorithm. The number of time steps processed in each parallel modal superposition operation is dynamically adjusted according to available GPU memory to balance memory safety and computational efficiency. The proposed algorithm demonstrates generality and scalability for complex rigid-flexible multibody systems with arbitrary topology. The effectiveness and computational efficiency of the algorithm were verified through two numerical examples. The algorithm achieves maximum speedup ratios of 7 and 42 for generalized coordinate calculation and modal superposition calculation, respectively. This provides an efficient GPU-based alternative for transient forced vibration analysis of rigid-flexible multibody systems.</p>