The simulation of fluids is an active research topic in computer science and is used in fields of medical engineering, astrophysics and computer graphics. The demands of the fields are quite different. Computer graphics prefer fast real time effects while the medical engineering requires realistic and robust methods. But an important factor for all fields is the efficiency of the simulation. The computation of such simulations is a time consuming task and requires optimized methods and implementations. One widely used method for fluid simulations is the Smoothed Particle Hydrodynamics method, short SPH, which divides fluids into discrete elements called particles. The interaction between particles is computed from the surrounding neighbors in the interaction sphere of each particle. Therefore the method performs a neighborhood search for each particle in the scene. The cost for the neighborhood search is a large part of the overall computation cost so that a fast neighborhood search implementation is required to increase the performance of the simulation. To achieve this, the neighborhood search must be performed in parallel. Since the introduction of the graphics processing unit (GPU) as a high performance computing tool, an implementation that targets the GPU is the best solution to utilize the benefits of a parallel algorithm.