A pipelined architecture for ray tracing

Ray tracing is a computer graphics rendering algorithm well known for the realistic images that it generates. Its primary drawback is the huge amount of computation required for even moderately complex scenes. Pipelined architectures have been used for many years to accelerate conventional scan conversion graphics systems. This thesis explores a pipelined architecture that implements the ray tracing algorithm. Various methods utilizing component parallelism, image parallelism, and object parallelism have been explored to accelerate ray tracing. Pipelining offers some advantages over these methods. Balancing the computational load between processing elements is easier. Only one copy of the database is required for each pipeline. It is also easier to implement the algorithm using specialized hardware.; A method of nonrecursive evaluation of the ray tracing algorithm is presented along with an architecture that implements this method. An implementation of this architecture called the RayPipe machine is also presented along with images generated by a software emulation of this machine. The software emulator is also used to provide performance estimates for a hardware implementation of the RayPipe machine. Complete chip and board schematics for the RayPipe machine are included.