How Many Iray Iterations or % Convergence Do You Find Necessary for a Good Photoreal(ish) Print?
Patroklos
Posts: 533
Clearly this will vary depending on the subject, how well lit, and just how real you want the render.
Onscreen I have never been able to see any difference between a 40% convergence and a 90%, indeed, to my eyes, often a 10% convergence produces a good render to view onscreen.
It would help if you could give a number of iterations or % convergence and give an idea of the type of render you are doing, particularly with regards to light levels.
I would be very interested on your observations on this.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
Iterations isn't a quality measure, like time it's a setting for how long you are prepared to wait on a first go-round. Convergence is broadly a quality setting, but some things will be more obviously problematic (noisy) than others and viewers will vary in how sensitive theya re to noise in particular areas so there probably aren't even vague guidelines - though I'm surprised that such low values work.
Thanks,
My understanding of iteration is that it is the point at which the file is updated, so 3 iterations is the file after 3 updates. Is that right? The size of the update would vary with what? The amount of RAM on the computer? If thats the case the question re; how many iterations is indeed meaningless as it would very from computer to computer.
I think I will run a little experiment, render a portrait at 1%, 5%, and 25% and see what I get.
Not bad at 1%? How much better will 95% be? It would certainly take a lot more time for a small improvement.
Presumably 1% is not this good on all types of render?
Much depends on the textures and the structure of the meshes. A rock with a "noisy" texture looks "done" way before a shiny, perfectly smooth sphere. In the first case the render noise quickly becomes comparable to the texture details, in the second case even a slighlty off pixel stands out from the crowd.
Shiny is out, grungy is in :lol:.
Merged stray post - if you are asked for a title it means you hit new Topic instead of Post reply.
Iterations is essentially sequences of calculations - the whole image is calculated, then the result is fed back in to the beginning of the calculation sequence and calculated, then the result is fed back in... Each "fed back in" is an iteration. The size is always one loop through the image calculating the light and surfaces.
'It depends.'
When I have a close-up with a lot of fine normals and displacement and complex transparency/refraction, I find it can take a lot of convergence for those tiny details to really come out.
When I have an otherwise complex scene where those details just aren't going to be obvious, then I find I can stop earlier, sometimes much earlier. (In one or two cases it wasn't even 1% and it was good enough)
My experience with Cycles is 200-300 samples for "easy" materials and easily 2000+ for "complex" one (glass/highly shiny/speckles). My initial sensation is that Iray is roughly the same (or slightly better).
There is also the route of rendering seperately the different passes and slightly smooth the noisier ones before compositing the lot.
When viewed at full resolution there seems to be a lot of noise (grain) in the image. It is especially noticeable in the eyes and to a lesser degree in the hair, but the entire image has a overall grainy look/feel. I'm not saying this is necessarily a bad thing if that is the type of look you want (as often people do prefer a bit more grainy feel to their images), but knowing it was done with Iray (or any progressive render engine) it is quite clear that it is "under cooked".
_____
Back to the original question (and I think the example posted by Patroklos demonstrates this), there really is no hard/fast rule. Some materials/lighting combinations resolve to an acceptable quality very quickly, others can take quite a bit longer (SSS, highly specular/reflective, and translucent materials like Crown Glass and Jade). If you don't want to watch the render to make sure it's "cooked" long enough, leaving the Convergence Ratio at 95% will typically ensure that your image will be of very good visual quality (if the max time is high enough to ensure the convergence ratio is reached).
How long you let you render run, and what convergence ratio you find visually acceptable is all a matter of personal taste/opinion. Some people set the convergence ratio to the maximum because they want the best possible image with no visual artifacts, while other may stop their renders at 1-5% convergence (then they often down sample the image to "filter out" some of the noise).
[edit]...nevermind, I found it.
I find 75% is often enough; I have a basic scene set to load on start or new at it has 81% as the cut-off. On rare occasions 95% isn't enough.
Generally, I find a hint of grain looks more photo-realistic; probably due to the fact that seeing that in photographs is normal, but as we get more used to digital images including photography and film then that expectation may reduce.
No...at least the movie part...because as part of the color grading/balancing process, most studios are adding 'film grain'. And I'm pretty sure, that unless you are making your own prints, from your digital pics, that most places that will print them are doing that too...probably automatically as part of the size/balance/get ready to print process.
I've got a scene that on 20 hours on Titan X with 12,000 iterations was still grainy. On other scenes, a few hundred iterations is enough. I've seen differences between 98% and 99% and other scenes where @70%, it's good enough. I find that shadows need a LOT more iterations. Also, any kind of fog, godlight effect or underwater difraction will need more iterations.
No...at least the movie part...because as part of the color grading/balancing process, most studios are adding 'film grain'. And I'm pretty sure, that unless you are making your own prints, from your digital pics, that most places that will print them are doing that too...probably automatically as part of the size/balance/get ready to print process.
That is good to know; it still may become less common, but may remain for a variety of reasons. Maybe ultra HD will have an affect and going forward the move to 8k and beyond; it could also be one of many options that print studios offer.
I don't want to predict the future; all history shows where that is concerned, is that the one predicting was way off at best. Speculating I like doing though. :)
@learning3d Look at this way : convergence means that all the light rays have been accounted for (computed if you will) In a scene where rays go all over the place, think a scene with cornered walls, some nice glass objects, some photoreal skin with lots of settings, it will take a lot longer to converge (compute) than a scene with uniform light, easy materials with a simple bounce (a plane surface) , simple spotlight with shadow.
HOWEVER, this does not mean you need 100% convergence all of the time. It might be your scene is filled with enough convergence that say the last 25% doesn't make a whole load of difference. In other scenes, some light might not reach some places enough, and even 99% doesn't look good.
So the amount of iterations is not that important, in the sense that X iterations always produce Y quality renders. It more or less depends on your scene. In time with experience, you probably can estimate your scene complexity and do some educated guesses.
I generally set convergence ratio to 99.5 or 99.95 percent.
I have had renders that actually hit 100% convergence that still had some grain. And it gets even better, something buggy was happening on that render (it was a GPU + CPU render, maybe that's what caused it..) and it went past that 100%, and still had some tiny bits of grain in shadowy areas, most notably in shadowy areas behind partial transparent objects. And the grain kept slowly disappearing in those areas, so I let it run to 107% before I canceled and saved.
There is this trick to simply double the dimensions of your render, then render to 40% convergence or so, and then size down the result image, but, well, that's for people themselves to decide whether that's an acceptable solution for them or not. I would probably end up doubling the dimensions, and still render to 100% before sizing down the result. Haven't bothered to try, but I bet the result would still be minutely different from the 40% convergence resized image.
The best answer is it depends.
Why? Because photo real can mean different things, and because you may not always want to do the same thing (please no
), and there may be different reasons and requirements for a certain look. Remember, photos have been manipulated for over a hundred years, yet all of them have been photo-real.
You'll get faster results if you spot render the results on the parts that have unacceptable grain, then composit the results.
Forcing Iray to work on all the image (even those parts that look done) will greatly slow it down.
... Sadly we can't chose areas for it to concentrate on - well, other than by spot rendering.
maybe someone can create a Render Brush
with a falloff that you rub over the render window after so many iterations to increase the iterations in those spots
moving the side and bottom sliders to see which parts of the image need touching up
There are a couple of render engines that allow for targetted resources, but forget which now.