In the automobile industry virtual prototypes are often used within the product development process. Here computer models of cars which are still under development are generated and analyzed to reduce the time and costs for building and testing real prototypes. In this context the technology of augmented reality offers a new perspective. In this paper we describe a system architecture for a PC-cluster consisting of AR nodes for the image processing and VR nodes for the parallel rendering. The AR nodes analyze the live video coming from two cameras and calculate the tracking data for the VR nodes. According to the tracking data the VR nodes render the 3D objects in parallel. The rendered images are sent back to the AR nodes where they are mixed with the life video. The proposed scalable architecture provides a very high image quality and rendering performance. This allows the superimposing of highly polygonal 3D models at high frame rates and excellent image quality.