There is a growing demand from the media industry, including computer games, virtual reality and simulation, for increasing realism in real-time for their computer generated images. Despite considerable advances in processing power and graphics hardware, increasing scene complexity means that it is still not possible to achieve high fidelity computer graphics in a reasonable, let alone real, time on a single computer. Cost prediction is a technique which acquires knowledge of computational complexity within the rendering pipeline as the computation progresses and then uses this to best allocate the available resources to achieve the highest perceptual quality of an image in a time constrained system. In this paper we describe a method of acquiring computational cost complexity knowledge within a high fidelity graphics environment. This cost map may be used in combination with other perceptually derived maps to control a selective renderer in order to achieve the best perceptual quality results for a user specified frame-rate.