Hi,
When rendering multiple light select RE (full mode), I found most of them have lots of noise in shadows. I guess it’s due to the threshold is cutting off based on the noise in the beauty(?).
Is there a way to force more samples into light(s) of a light select RE to get cleaner light select RE? I have tried enabling “Use local subdivs” in Render Settings and increase light’s subdivs, but it doesn’t seem to help in my tests.
beauty light selectThe noise may be coming from something else that the light. Especially if the light select is in FULL mode, then it might be GI noise, or reflections, refractions. See if adding a reflectoin, refraction and GI render element can help you identify the objects, materials and source of the noise. Then you can either use the local subdivs, or (in case it’s tied to an object/material and is not everywhere in the scene) use vrayObjectProperties and increase the subdivs mult just for those objects - this will work even if the local subdivs are not used.
@yolov , ah…I’ve totally forgottenten to check the source of noise from other render elements. Thanks for the reminder and the tip of using vrayObjectProperties. Will try that too.
Just a hint - with the subdivs mult in vrayObjectProperties you can raise the shading subdivs when “use local subdivs” is enabled. When it is disabled, the VOP subdvis mult raises the shading rate, which makes it quite useful and saves you the pain of switching to local subdivs. I personally use this little trick quite a lot.
@yolov I did a couple tests and the trick works great so far. I keep “use local subdivs” disabled, and just raise VOP subdivs mult to like 12 for objects that are part of that VOP. Since the noise only appears on certain objects, the approach works efficiently.
However, I’m a bit confused by your last comment. So:
1) VOP subdivs mult raise “shading subdivs” when “use local subdivs” is disabled
2) VOP subdivs mult raise "shading rate " when "use local subdivs " is enabled
What’s the difference between raising shading subdivs and shading rate?
Well, when you have “use local subdivs” - you use the material/light’s subdivs to control the number of secondary samples for them. When you have “use local subdivs” off, the min shading rate parameter globally controls the number of secondary samples. So in the one case, the VOP mult would raise what would be the subdivs parameter, and in the other - what would be the shading rate paramter. That’s all I meant.
@yolov , thanks for the further explanation. So for an object, if there is noise in reflection RE, what is the difference between a) “use local subdivs” on and doubling the subdivs of the associated material and b) “use local subdivs” off and set VOP subdivs mult to 2?
Probably:
a) would require re-adjust all other local subdivs, if you’re switching from not using local subdivs to using them just for this reason
b) might need more than doubling, i.e. you might need mult>2
@yolov I should be more specific on my question. 
In terms of sampling, what is the difference between a) and b)? If a material is only assigned to an object in the scene, and we found noise in reflection RE for that object, is there a difference between applying a) or b) approach to get rid of the noise?
Is there a simpler way to understand mathematically difference between doubling material subdivs vs. doubling the min shading rate? What I can find in Docs is that increasing shading rate means rays will be budgeted for shading effects and less about primary rays. It’s a bit vague whereas doubling material subdivs is self-explanatory.
Verbose mode on
The difference is in the way the number of samples will be determined. When using local subdivs, the image sampler’s max subdivs value is taken into account. When a light (for example) uses 8 subdivs and the image sampler is set to 1/4, the samples for the light will be 8 subdivs = 8*8=64 samples but divided by the sapmler’s 4 subdivs = 4*4 = 16 samples, so 64/16=4 samples for the light. Increase the light’s subdivs to 16 and you get 16*16=256/16=16 samples. Increase it to 32 and you get 1024/16=64 samples.
Now why does V-Ray divide the local subdivs by the sampler’s subdivs, well - to prevent taking way too much samples (that was the original design) and slowing down the render too much.
Then came the “divide shading subdivs” checkbox. Disable it and all that maths will not happen, so essentially with “divide shading subdivs” = off, and for the same 32 light subdivs you’ll get 1024 samples.
In the above setups the min shading rate will determine the ratio between AA and shading samples per AA rays, essentially to bias the sampling towards more AA or towards more shading samples. I would probably need to run some tests to get you some exact figures about this, cause I’m mostly making this up (I haven’t really tested such a setup for years since the min shading rate = disable local subdivs options came to V-Ray).
The beauty of setup simplicity, though, comes when “use local subdivs” is off and you only play with the image sampler’s subdivs and the min shading rate.
In brief, it would mean for image sapmler 1/4 and shading rate 1 >> 1 shading sample for each AA ray, i.e. for the pixels where the sampler has reached its 4 subdivs or (4*4=)16 samples, there will be 16 shading samples for each shading effect (16 for lights, 16 for reflections, etc). Increase the shading rate to 6 (the default) and you get 6 shading samples per AA ray, or for the pixel where the image sampler made the max 4 subdivs, you get 96 shading samples per shading effect per pixel (4*4=16 aa samples and for each of those you get 6 shading samples).
I should note that the above examples rely on no adaptivity, i.e. I’m relying on a theoretical case where with 1/4 image sampler subdivs the image sampler actually manages to make the max 4 subdivs (16 AA samples) per pixel.
With all of this said, in a case where you want to clean reflection noise coming from a specific object, what matters most is your initial setup.
* If you initially have setup V-Ray to use local subdivs and divide the shading subdivs to the sampler’s subdivs, then you will have to take that into account and when increasing the object’s reflection subdivs keep in mind that they will be divided by whatever your sampler’s max value is. So you might need to not simply double them, but triple, quadruple them and so on. Bumping up the subdivs mult in the VOP node (vrayObjectProperties) will just multiply them too to the same effect.
* If you’re initially using local subdivs, but not dividing them to the sampler’s subdivs, then no math is involved - just up the reflection subdivs or up the VOP’s subdivs mult. That will simply mean more samples.
* if you’re using the defaults - use local subdivs = off, then the VOP’s subdivs mult will multiply the min shading rate value for the pixels that your object takes up. It will result in more samples for not only reflections, but all shading effects and probably make V-Ray oversample those pixels, but it is really simple to do. Also, V-Ray has become quite smart in recent years, so I’d expect it will know what to do (I have been consistently failing to trick it into behaving in a less adaptive way than it is by default).
This might shed some light onto why the new default is better (I think it is, use local subdivs = off and tweak min shading rate) since less math is involved and less things to consider when tweaking the render settings.
And some info to back this up: Chaos Docs
More specifically where the Min shading rate and the Divide shading subdivs are mentioned.
@yolov Hey Alex, finally got some time to re-read your awesome post to break down how the sampling works in those situations. It’s refreshing to me as I found myself always learn something new when I read it.
The forum is really valuable because of posts like this.
I agree that the defaults is way the go to begin with. It’s simpler to set up the scene and can still tweak to get rid of noise before users get under the hood to turn on use local subdivs and turn off “divide shading subdivs”.
Thanks again!
oh god please not this local sub div thing, i thought we got rid of this
It’s disabled by default and we recommend keeping it this way. It’s there to be enabled only if needed.
