Thanks for the detailed info. Yes, I can agree that RPS setup time is a pain in the ass, but it's so customizable. I do agree that as computing gets faster, shops will move to raytracers, as man power is more expensive than more computers.
As far as MB, DOF and DISP being as fast as a reyes renderer...hmmm...it seems hard to believe, especially since it goes against the logic of a raytracer vs a reyes renderer. I believe you... I just wonder how it's doing it...
It's so damn hard to understand any of the technology when it's not available to anyone but big studios.
Arnold's not just an executable you chuck geometry and textures at and get an EXR back. It's a complete C++ API as well. So you can use it to write your own raytracer / raycaster, and integrate it completely with your own tools and pipeline, and just let the Arnold library do the hard work behind the scenes. You can write your own integrator, accelleration structures, etc, etc, so it's pretty much as flexible as RPS... That's why the big studios are going for it (and why SA are not selling it normally).
The only other thing I know of it's not as fast with compared to PRMan is deformation motion blur (as opposed to just stocastically sampled transform of the primitive) - I think PRMan still wins here because they can do triangle culling which a brute force GI tracer can't do. But it's seriously fast.
Even more useful information. Thanks again.
If you're at siggraph let me know...I owe too many beers in this forum, so one more won't hurt...