Abstract: Simulating large-scale fluid scenes in real-time is of great value in both research and application.To achieve this goal,we present an integrated algorithm for fluid scene simulation.A new function of fineness is proposed to make decision in our non-uniform particle re-sampling process to both reduce the number of particles in need of simulation and enhance simulation speed.We also propose a novel 3D spatial grid partition algorithm and parallel radix sort algorithm to increase speed for searching neighboring particles and interacting with boundaries;We use the new NVIDIA^® Compute Unified Device Architecture(CUDA)to compute SPH entirely on GPU,which makes full use of the high parallelism and programmability of GPU to simulate fluid in real-time using SPH method.Experiments show that the method proposed can be used to simulate fluid scene in real-time with satisfactory effect,and the computation speed increases up to more than two orders of magnitude in comparison with the existing CPU SPH methods.More detail effects than other GPU SPH methods can be generated.