3D Ray Tracing

I need to read the api again, but to me the api did not seem to require improved modelling, triangulation or uv mapping its main focus appears to be that of providing more efficient culling so that only the rays that will appear in a scene actually need to be traced, thus making ray tracing more affordable.

The way I think of it that old school rasterization works from the viewer towards the object and culls the z-map as it goes. The M$ api ray traces only the visible pixels back from the object to the viewer.  This incidentally ought to automatically comprehend occlusion effects – the sort of fuzzy edges that objects can have. It should also automatically generate shadow maps.

My guess would be that its first impact will be on very much improved in-game graphics.

Just a follow-up comment, while of course your comment about texture detail is correct it really only applies to background LoD work where probably an average normal map is burn into the textures, but for foreground action work that HAS to be done with polygons and where necessary distorted by skeletons and whatever process is used to transfer the result to the game engine (e.g. Maya->Cry Engine)

Dan, the problem with game RT rendering is that it has to make compromises to meet the insane target of an unscripted  rendered scene in less than 0.017 seconds on a high mid-range CPU and GPU . A typical Blender or other cut-scene render can take as long as necessary using as large a render farm as it needs and just save out the finished frame.

The frame rate is the killer that drives reality compromises as a game that achieves <60fps gets very critical review comments. Level of Detail (LoD) modelling (polygon count/object goes down with distance) is a typical compromise, but you will still hear complaints about ‘popping’ as objects move from one LoD to another. There are many other artificial (unnatural) artifices that the developers have to use to reduce render time. Even then a glance at any face in any RT scene shows up as glaringly unreal – as you know flesh is very difficult to render even in pre-renders and RT just cannot afford the fps to do anything except teeter on the edge of the unreal valley.

If Moore’s law holds maybe another 10 years will see some of these artifices becoming unnecessary, and it will become more common to ease the burden on artists using natural scans of scenes and objects as shown in the demo you linked. I’d also see some of these techniques spilling back into the world of Blender as artistic time constraints also restrict time spent on design and creativity. I think it could be 20 years before the difference for pre-renders becomes virtually identical to RT renders, but that of course does not allow for the impact of 8K screens or 48bit true colour!

I suspect it is a ‘competition’ in which pre-rendering will always be slightly ahead.