a general question out of curiosity
Drogo Nazhur
Posts: 1,079
in The Commons
Why do some scenes finish rendering after a few iteriations while other scenes hit a thousand iterations and are still on 0% complete?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Because the lights, materials, and render settings affect how quickly a render will converge. If you render a scene with no geometry and no lights, it will be the fastest to converge since there are no other paths than black pixels. The more paths lights can possibly take, the more samples are needed for pixels to converge. You can see this by rendering a scene with many reflective objects vs the same objects with rough/low specular materials.
To be more precise, there is a difference between Iray rendering times and 3dlight rendering times.
For Iray, the more light you have in a scene the faster it will render (reflective surfaces still will add iterations)
3dlight can handle darker scenes better, for this what RobotHeadArt said is 100 % true
Remember for Iray there is a difference between more light and more lights. A single bright light will generally render faster than a large number of lower intensity lights. In particular high poly objects should not be used as mesh lights since each polygon is a different light.
Convergeance is Iray looking at how much the latest batch of iterations has affected each pixel, and if it's little enough, it assumes that more iterations are likely to do the same, and the pixel is good enough to call "done". Then, when enough of the pixels are "done", then it deems the entire image to be complete. Exactly what your quality and convergenace ratio settings are defines how discerning Iray is in this assessment of each of these things.
Scenes with simpler light paths (i.e. fewer bounces, such as will often happen with an exterior scene where light quickly bounces to an HDRI light source*) will generally require fewer iterations for each pixel to reach the convergeance point.
~~~~~
*It should be noted that in Iray, light paths start at the camera, then bounce until they either reach the maximum allowed path length (which is number of bounces, not distance) or find a light. While this sounds backwards, this is common practice for ray tracing, as scenes usually have one camera but multiple light sources (or at least a much larger area of light source than lens), so it's much more efficient to take the approach that assumes that every ray path you care about ends at the lens, rather than calculating a load of paths for a light way off in the distance that are extremely unlikely to randomly bounce into the lens.
This actually means that the main factor in reducing iteration time is larger lights as the ray paths are more likely to find one in fewer bounces, whereas brightness of a light source is fairly irrelevant as far as the algorithm is concerned.
(Lots of people will say "Render bright! Darken in post!" but no, that just doesn't add up. Iray is calculating in floating point values; the accuracy of the calculations is unaffected by the order of magnitude of the numbers used).
that actually explains a lot of things, thanks @Matt_Castle
As Matt Castle says, it is the odds of hitting a light that matter - so a single, bright light will not be as good as several lesser lights. However, the complexity of the light geometry does matter - as I understand it - so you want to keep the light objects, if any, as simple as possible.
Because the iray renderer is possessed by demons.
OK, OK, I kid, but some days that seems like the most plausible explanation.
Somewhere in The Missing Manual For Daz Studio there is a section which explains things like how you can hit a point with photometric lights where adding more light to the scene no longer improves render time and quality because the renderer is now doing so many photometric light-bouncing calculations that it's bogging down; explains that sometimes it's better to be all-or-nothing with photometrics because photometrics and mesh lights (emitters) don't play well with each other in many situations and putting even ONE photometric light into a scene immediately turns on "got to think in photometric mode" in the renderer brain and changes everything; that for emitters the density/complexity of the mesh is the key factor; if you make your ceiling-light surface or a plane primitive emissive, that's not a lot of mesh cells, but if you make your human figure emissive and you had an HD mesh, you've not made one emitter, you've made hundreds, and your render time is about to do some horrible things. (Ask me, who makes a lot of glowy humans, how I know this.)
No one told me any of this three or so years ago when I started, nor was there any good place to learn any of it. There still isn't.
I don't think most of those would make the manual for any application - though for Iray you should be looking at the nVidia docs anyway.
I am told that the interpretation of what you are seeing is incorrect - I'm afraid it is not soemthing of which I have first-hand experience to make any better explanations.
Which part? The photometrics-not-always-getting-along-with-emitters part? Very possibly! I'm always willing to accept "I'm wrong" as the most likely explanation :P
I have never tried to go systematically into it; I don't have the spare afternoon to try "OK, let's see what this light does with this light in a controlled situation" over and over in various debugging combinations. I just know what I get sometimes, and I have a reasonably reliable way to fix it, which is to not use photometrics in that scene at all. What's really happening, and why that should make a difference, is a different story.
Maybe the next time I get a scene that's misbehaving in this way I'll remember to stop and do some more checking ... I also can't predict when it will happen, so it might be a while.
Well, the idea that there is some incompatibility or different coding - I was not disputing the experience of problems.
It does occur to me that if the photometric lights are quite bright they are going to have a noticeable effect on the final pixel value even with the fill lights, so that will slow convergence in proportion to the time their contribution takes to settle - and making them brighter will make that even more apparent.