High Performance Computing For Image Synthesis

For more than 20 years, several research works have been carried out to design algorithms for image synthesis able to produce photorealistic images. To reach this level of perfection, it is necessary to use both a geometrical model which accurately represents an existing scene to be rendered and a light propagation model which simulates the ligh propagation into the environment. These two requirements imply the use of high performance computers which provide both a huge amount of memory to store the geometrical model and fast processing elements for the computation of the light propagation model. Moreover, parallel computing is the only available technology which satisfies these two requirements. Since 1985, several technology tracks have been investigated to design efficient parallel computers. This variety forced designers of parallel algorithms for image synthesis to study several strategies. We will present these different parallelization strategies for the two well known computer graphics techniques: ray-tracing and radiosity. This presentation will focus on two difficult problems to be solved: exploitation of data locality and load balancing.