Wave-induced disasters usually lead to casualties and substantial material losses, and seriously impact the local economy and ecological environment. As a meshless particle method, smoothed particle hydrodynamics (SPH) has been widely used in wave propagation, breaking and FSI problems, and is becoming an effective method to simulate wave-induced disasters. In this work, a matrix-free incompressible smoothed particle hydrodynamics (ISPH) is proposed for large-scale simulation of wave-induced disasters with GPU acceleration. The pressure-Poisson equation (PPE) is solved by Jacobi iteration with an adaptive relaxation method, where minimal residual relaxation (MRR) and orthogonal projection relaxation (OPR) are adapted to obtain the coefficient of relaxation adaptively. On the one hand, the present method accelerates the rate of convergence. On the other hand, the matrix of PPE is not required by the present method, which avoids a large amount of cache and thus greatly improves the capacity for large-scale simulation by GPU. The dam break, water entry and liquid sloshing are used to validated the GPU-accelerated matrix-free ISPH method. Finally, the wave-induced disasters, i.e., flooded city and waves impact offshore structure, are numerically investigated by the present method with more than ten million particles. The results demonstrate that the GPU-accelerated matrix-free ISPH provides a competitive method for wave-induced disasters with a fast, accurate, large-scale simulation. In addition, some valuable findings of wave-induced disasters are also reported.