Description
Ray-tracing is often suggested as a problem which is well suited for execution on multiprocessors. It is characterized by having abundant parallelism, a very small sequential part, and aggravatingly long run-times. For simple will behaved scenes, linear speedup is easily achievable. However for realistic scenes which are typically both complex and non-uniformly distributed, parallel ray-tracing is a challenging problem. This thesis evaluates and compares implementations of a sophisticated ray-tracer on both shared memory and distributed memory machines.