Ray Tracing In General

nobond

Newcomer
HI, the question might be vague

Just generally, if we completely ignore the cost, is it true that ray tracing approach is superior than the raster approach ? Or is there anything ray tracing is not good at ?
 
Superior by what metric?

What it's not good at is dynamic scenes where triangles change positions. Simple non-skinned models (a'la quake3-HL1 era) are doable with hierarchies, anything more complex will be a huge hassle.
 
If costs are out I'd just let a gazillion photons loose in the scene and simulate actual light instead of faking it with ray tracing :)
 
Also Ray Tracey's blog is quite nice place to see all sorts of cpu/gpu realtime raytracers.
http://raytracey.blogspot.com/

Brigade seems to have flooded first couple of pages, but there is all kinds of tracers in older pages.
 
Last edited by a moderator:
If costs are out I'd just let a gazillion photons loose in the scene and simulate actual light instead of faking it with ray tracing :)
Why stop there? Some years ago (more 15 I suspect) there was a paper that did rendering by propogating wavefronts!!!!!
 
Raytracing has the same problem it has always had, it is not good at being fast for primary rays and (approximate) shadow rendering for scenes without lots of subpixel polygons... and that's still pretty much what we are doing in real time as far as visibility goes. Once we go down the path of subpixel geometry and non precomputed GI the time will have come for raytracing.
 
Last edited by a moderator:
Raytracing has the same problem it has always had, it is not good at being fast for primary rays and (approximate) shadow rendering for scenes without lots of subpixel polygons... and that's still pretty much what we are doing in real time as far as visibility goes. Once we go down the path of subpixel geometry and non precomputed GI the time will have come for raytracing.
A rasterizer optimized for very small primitives looks a lot like a ray tracer minus walking over the acceleration structure (for instance one could design a rasterizer that handles 2 small primitives that share an edge). Even in that case I am not sure I'd go with a ray tracer for primary rays anyway (certainly not if the model is dynamically tessellated to sub-pixel triangles).
 
There has been a lot of work on micro triangle rendering in recent times. None of it ray tracing related.

Reyes has been doing stochastic rasterization with sub pixel triangles for 25 years now.
 
Back
Top