A Massively Parallel Approach for the Design of a Raytracing Oriented Architecture

Abstract

Solving time critical problems requires a computing power of an order of magnitude greater than todays available conventional computers. The use of massively parallel architectures appears to be an attractive and effective way towards the required performances. The ray tracing technique is known as the best synthesis method for the construction of realistic images but also as the most time consuming. Computation time of several hours per image on a conventional mainframe is usual. Fortunately, this technique exhibits a huge amount of potential parallelism and therefore massively parallel architectures fit well and straightforwardly. This paper presents an efficient implementation of the ray tracing algorithm on a dedicated network of transputers. The INMOS's transputers are a family of monochip processors specially designed for parallel, asynchronous architectures without shared memory.