<p>This study presents a scalable computational framework for the direct simulation of incompressible fluid–structure interaction (FSI) involving rigid bodies, combining a multigrid-accelerated projection method with a sharp-interface immersed boundary formulation on distributed GPU platforms. While both methodologies are well established individually, their systematic integration within a fully parallelized environment for large-scale FSI simulations remains limited in the literature. The proposed solver employs a fractional-step projection scheme with third-order total-variation-diminishing Runge–Kutta time integration and incorporates a dual-time-stepping strategy to enhance the convergence of the pressure correction. Fluid–solid coupling is enforced explicitly using a velocity-interpolation-based immersed boundary forcing method, enabling sharp interface representation and accurate satisfaction of the no-slip condition without introducing excessive numerical diffusion. The numerical framework is validated through a series of canonical benchmark problems, including three-dimensional lid-driven cavity flows with internal obstacles, oscillating rigid bodies, and sedimentation of single and paired particles. Complex interaction phenomena, such as vortex shedding and the drafting–kissing–tumbling sequence, are accurately reproduced. Large-scale simulations involving the gravitational settling of up to 20,000 fully resolved particles are further performed, demonstrating the robustness, scalability, and parallel efficiency of the proposed approach. The results indicate that the present framework provides a reliable and efficient tool for large-scale FSI simulations in computational mechanics.</p>