Ray tracing is a widely accepted technique for generating realistic images of artificial models. One of its major drawbacks is the time needed to compute an image. For this reason parallel processing is often used to accelerate image generation. This report presents a ray tracer in which a hybrid scheduling technique is employed. Scheduling consists of a data parallel and a demand driven component. The latter uses the Pyra clip method to efficiently compute intersections for a large number of rays and at the same time keeping communication requirements within bounds. Which rays to compute in a demand driven manner and which using the data parallel algorithm, is decided by the amount of coherence between rays. Single rays are traced by the data parallel algorithm and coherent rays are processed by the demand driven algorithm. The result is a scalable ray tracing algorithm capable of rendering images of large models.