We are developing a new-generation multi-node multi-GPU solid-fluid simulation platform which involves dynamic coupling among air, liquid and solids, where many iso-thermal multi-phase phenomena can be simultaneously well captured. Unlike the traditional approaches for solving fluid flows by discretizing the Navier-Stokes equations with either structured or unstructured meshes, our simulation is motivated from the statistical mechanics, and based on the kinetic formulations. With such an approach, the originally complex fluid flow equations are simplified into pure algebraic equations with local updating rules, which can be directly iterated for fluid flow simulations, without solving any pressure equation or global linear systems. This dramatically increases the system efficiency and scalability. With our adaptive central-moment relaxation scheme and particular force discretization, our simulation platform can achieve sufficiently high accuracy and stability. Due to local dynamics in our solver, the sample placement can be quite arbitrary and the boundary treatment is also easy. Thus, the complexity of geometric structures is no longer a problem, which enables our platform to easily solve for fluid flows involving complex static/dynamic solids. The whole simulation platform is realized with the goal for high-resolution large-scale fluid flow simulations, with applications in movie industry, medical diagnosis, industrial product design, as well as intelligent robotic training, etc..