Hello, When using SIMD optimizations in ray tracing, I would like to know what is the most efficient method, optimizing the code to use less instructions (with SIMD) or by coding the code to use mutliple rays at the same time (same number of instructions but having them executing on multiple rays). The paper Interactive Rendering with Coherent Ray Tracing uses packets of rays and gets high speed improovements, but there's no way to optimize this way for reflected and refracted rays (and some shadow rays). Therefore I would like to know which method is more efficient if you consider the secondary rays. Another thing I have trouble with is some code cache optimizations. The ray tracer in the paper only uses triangles as primitives and I consider this (personal opinion) bad use of one of the ray tracing's features (perfect parametric surfaces). But when you put various types of primitives there is a consistent change of code in the cache due to the swapping of each primitive's functions to execute them. So one solution would be to reorder the primitive list before rendering and group them by primitives. Can this be a significant gain in performance or multiple primitive types is just a performance draw-back? Thank you, JVFF EDIT: Just a revision with some corrections [Edited by - jvff on October 10, 2004 8:08:54 AM]