Iray rendering too quickly.
...OK this may sound an odd complaint, especially for someone still stuck with CPU based rendering, However I am noticing with one scene I am working on that Iray rendering completes in what I could consider "record time" for my 4 year old system. I am using the 4.9.166 beta and the scene is my Leela character on as stage at hte piano with two ghost lights and two photometric spotlights.
When I start the process the progress pane counts off iterations one by one as if I am using the GPU (which I am not since it has only 1 GB VRAM) By the 32nd iteration convergence jumps from 0% to between 79% and 83%, by the 50th or so to 88%, the 92nd, 94%, 99% at 122, and completes after only 150 iterations (about 9 - 10 min). The issue is after completion, the image is still noisy (even though I have both the Noise and Firefly filters turned on). Not sure why this is occurring as I am using my standard Progressive parameters of 15,000 iterations, 8 hours maximum time, and 99.9% Convergence Ratio. The scene uses two Photometric spots and two low intensity ghost lights along with the old Opera house stage, one of B999's Backdrops Made Easy (with floor hidden), A Grand Piano model and the Leela G3F character.
Test renders of just the character with my photo studio setup take longer to complete (about three to four times longer) yet are better lit with less in the way of polygons and textures. It makes no sense as usually with Iray, darker lit scenes hould take longer Using just the "Scene Only option as any of the other ones brings in the default HDRI/Sun-Sky settings.
Comments
I recommend you turn Quality "off".
Now Max Time and Max Samples control the length of the render.
I prefer to set Max Time to "0" effectively turning it off and control the render using Max Samples. If I'm happy with the render before the Max Samples is reached, I stop the render. If I'm not happy with the render after it's reached the Max Samples, I just increase the number via the controls on the left of the render image and hit resume.
Unfortunately, you can't turn off Max Samples, so if you want to control the render via Max Time, set the Max Samples to a very high number.
I suspect it takes slightly less time with Quality off because the software doesn't have to calculate convergence, but it's probably neglible. Though you might see a difference with your older machine and cpu only renders.
ETA: I also, usually, turn on the Architectural Sampler. Oddly enough, it can shorten the render time or lengthen it, depending on the scene. But it renders the entire image at the same rate. With it off, the render engine gives priority to light areas. That allows darker areas to still be noisy/grainy when the rest of the image looks fine.
Maybe there is too much bright light?
Thank you so much for this information. I'm having the exact same problem using an emissive plane in an enclosed area. I will definitely try this and see if I can get a less grainy render.
...light settings are pretty minimal. The ghost lights are at 5,000 cd/m ^2 (default) and spotlights at 10,000 lumens with tone map settings of ISO 100, shutter speed: 1/15, F/Stop; f7. As I mentioned the tests which I did of just the character using my a standard studio light set, that is a lot brighter, takes much longer to render usually in the 1,000s of iterations rather than just 150.
Spent the better part of the other evening and next morning (until about 03:00) messing with difffernt light and render settings and still finished with a lot of noise.
..thank you. Giving that a go. Also set the emissive planes to cd/ft^2 which is the inverse square rule.
You are both welcome. I'm more than happy to let an image render as long as necessary. Doesn't take so long now, with the beast, but when I was cpu-only, I'd let an image render for three days or more to get a decent quality.
I'm not familiar with using the inverse square rule on an emissive plane. I'll have to try that.
...well so far at 660 iterations and still going When I was sill using Reality/Lux I had to just trust my eyes as to whether or not it looked clean as Lux will render to infinity if you just let it go. (although it was severely glacial in that even after 15 hours it looked like it was maybe about 20% - 30% converged at best.
Need to come up with 700$ for a 1080 Ti as that would pretty much render most of my scenes with little trouble.
You might have resolved it by now, but whenever you have perplexing light/render issues with Iray, a good course of action is to restart your PC. Sometimes things get stuck in a loop, or something.
If things don't improve with this particular scene, you should be able to force Iray to do a deeper convergence with the Rendering Quality settings. The default is 1; higher numbers increase the threshold of how Iray considers pixels to be converged.
The other other settings -- time, samples, percentage ratio -- are stop-at values and don't actually alter the characteristics of the render. They simple allow the render to process for a longer or shorter time, and sometimes, the convergence estimator in Iray is way off. I've had scenes where there was no apparent unconverged pixels, and the thing was still stuck at some low value. (That problem resolved itself with a restart, BTW.)
...I restarted the system since the other evening and it still was finshing with a minimal number of iterations. Turning off Render Quality seems to have done the trick and I'm getting much better results albiet rather slowly (as I would expect with a low light scene).
Yeah, conversely, also pushing the quality up will also force it to render longer. If you set quality to 4 or more you'll substantially increase your render times and it won't turn off before the noise is gone. I hadn't tried ever tried setting quality to zero before. I might have to give that a try some time.
Just out of curiosity did you have the Head Lamp on? I ran into this the other night. I set up the scene was using some new lights, I had tweaked the lights to make it rather dark. When I rendered it, it was finished in 4 minutes but had a lot of firefly's in the highlight areas and noise in the dark areas. I usually render at quality level 3, so most of my renders take 30min to an hour or so. I checked my render settings several times and re-rendered, again 4 minutes same issues. Finally checked the camera settings saw the head lamp was on. Turned it off. Render took an hour +, and came out with no issues.
...nope, always remember to turn that off when rendering, otherwise it looks like a photo taken with a flashbulb.
I used to increase the quality settings to force Iray to render longer. I'd set Max Time to "0" and Max Samples to some high value. but it's "inexact"... 99.99% convergence can be some low value, as Kyoto Kid was having trouble with, or it can seem to go on forever. I've found 2500 samples to be sufficient for most of my renders, and I don't have to guess what Quality value will allow my image to reach that goal. Now I have DS set up to open a default file with my favorite render settings already in place.
I haven't read anything to convince me a render using Quality and Convergence will provide a better image, and I prefer the convenience and control of using Max Samples as the sole cut-off parameter, especially with images using low light. But that's just me. I like to say I'm a recovering perfectionist... but to say "recovering" might be a bit of a stretch.
Ohh It never occurred to me to turn of the quality setting.
I'm definitely going to try this, thank you for posting :)
@kyoto kid " Also set the emissive planes to cd/ft^2 which is the inverse square rule"
Changing units doesn't "invoke" the inverse square law for light. The Inverse Square Law is an inherent property of light which doesn't depend on the units of output or the value of the output.
The Inverse square law describes light falloff and states that the amount of light per unit area will lessen (fall off) by the inverse square of the distance from the source, i.e. 1/4 at 2 meters, 1/9 at three meters, 1/16 at four meters and so on.
When you changed your emissive to cd/ft2 from cd/m2 all you did was make the emissive light brighter (more total output).
Probably changed it from Lumens to cd/ft^2.....which DOES alter the calculation (lumens are total light output, cd/ft^2 is light output per square unit at distance.) Though it doesn't alter the 'falloff' which is an inherent property of spreading light.
In the Iray parameters, using Cd/ft^2 is based on original steradian measures, for a circle of 1 square foot at a distance of 1 foot (assuming a point source.) Lumens utilize the actual surface area, and divide photon groups across them (probably with linear interpolation across normals.) Using Cd/ft^2 or Cd/m^2 allows you to not have to use such huge values as needed for Lumens.
(Personally, I think the calculations in Iray are not correct, as using the numbers for Watts seems to produce WAY too little light. Somewhere in there, probably in the 'watts' definition in Iray, there's something off....)
...well the lighting looked better with the cd?ft^2 setting
After 7 hours and 50 min (6,465 iterations) this is what I got (the covergance rate was only 48%).
Another option to try is very simple - adjust the camera position ever so slightly. I've had the "too quick" issue in the past, and adjusting the camera position usually fixes it, even if the new camera angle is only slightly different than the old.
@hphoenix "Probably changed it from Lumens to cd/ft^2....."
The OP said 5000 cd/m^2.
"(Personally, I think the calculations in Iray are not correct, as using the numbers for Watts seems to produce WAY too little light. Somewhere in there, probably in the 'watts' definition in Iray, there's something off....)"
When using emissives, luminous efficacy comes into play. The default may not be correct for the lighting value (product) one expects. The default is about right for a 60W Incandescent,
...well adjusted the lighting angles and intensity a bit and let it cook overnight (total of 11142 iterations). Her shirt still looks grainy but I thik it is because I forgot to adjust the tiling of the tecture to a smaller scale.
Looking good.
I think the grain on the shirt could easily be mistaken for fabric texture. I wouldn't have thought it noise/grain if you hadn't said anything.
...againm it probably is because I forgot to adjust the tiling (used Mec4D's PBR shaders vol 2).