Am I right to assume this isn’t possible at the moment? I found some older threads through the years.
I’d like to be able to set image sampler subdivisions per object or material. If you have a visually noisy texture or object, V-Ray can spend a lot of time trying to clean it up despite the noise being not noticable to the human eye. In the object V-Ray properties, there’s a secondary ray multiplier, but that has very little effect.
The subdivs multiplier as you mentioned should work as intended. So you would lower this value if you want less sampling in that area. This doesn’t have the intended effect?
Thanks for your reply!
No, since that’s only for the secondary rays. The main sampling is still done according to the global values set in the render settings.
A yes, sorry it’s saturday, afraid you can’t do that anymore, this functionality has been deprecated as the algorithm should be smart enough AFAIK, can you post an example like a crop or something,? There are a few other ways to speed things up but it depends on the scene / material / etc.
Also check this out:https://www.youtube.com/watch?v=pKz34yrDxJE
Not sure if this might apply to your project as this only works for still images. Just tested this out and it’s actually quite good. I have less noise on 8K resolution (with threshold set to 0.2!) than rendering with the defaults at 0.01 on 4K resolution. I was always anti-denoiser but this clears most of the issues I had before.
This is very interesting, but why do you say it only works with still images? Is it a temporal inconsistency thing maybe?
I have a nightmare render at the moment, of 10k HD frames, each at 10mins (80 days 24/7 rendering), so I was immediately thinking I could cut the overall time of the remaining sequences by using this method.
Yeah, I haven’t tried it with animations, but afaik the Intel denoiser doesn’t have temportal consistency which will lead to possible flickering. But depending on the situation you might get away with it and maybe even try some deflickering with Topaz Video AI or similar.
edit: to not hijack the thread I made a different 1 to discuss this further and maybe post some examples: Lower rendertimes "hack" - Denoiser trick - V-Ray for 3ds Max :: General - Chaos Forum
I don’t have VideoAI so I’ll wait for the current sequence to finish and do some quick tests with defaults.
It could still save me a chunk of time if it works out, so thanks a lot for posting it 
Why it isn’t common knowledge is a little irritating, as clearly Chaos people knew of this approach, as was stated in the video.
I have used the intel denoiser with animations with very good results (always blending with the original some). It depends on the shot. For depth of field noise cleanup the intel was actually pretty amazing. Was doing an underwater (under cola) shot of fizzy bubbles and ended up with crazy high sampling settings like 0.05 or 0.08 and it looked pretty nice.
Worth a try, but it’s all so scene dependent. I wish you could render all the settings for all denoisers at once (can you do that already somehow?).
I decided to stick with default Vray one for now.
I reckoned that because my shots all comprise single camera moves, going from wide entire scene view zooming to close ups with animated stuff, that it wasn’t worth the risk and was just too lengthy a test to do.
Yes, I think you can add a separate denoiser element for each type, then just run the denoise later to compare. I really should have done that
for this project but given it’s 10k frames I don’t really have time to waste testing/processing all that data.
When I do use the denoiser I render to EXR with the denoiser elements and then set up frame dependent jobs to run the denoiser usually on older/slower machines while the render is still going. That way both versions are ready around the same time as the main render.
Good Luck!
Hehe, yeah I wish I had more than the one machine but sadly I don’t, so all I can do is wait.
I had 6 sequences to render and I’m on the 5th now. Luckily, good planning wins out and so so far, I’ve
have no issues at all really, despite the relative complexity of each one.
The only hiccup was a Windows update that stopped Max just as I’d gone to sleep, losing me 8+ hours.
This project has taken almost a year to get to this point, so I really cannot wait until it’s done and forgotten about 
This changed in V-Ray 3.3
With the max AA high enough, the image will automatically converge across the screen, no matter the scene contents.
For a quicker render it’s enough to raise noise threshold.
Use a sampleRate RE to know where the image has converged or where it hasn’t (red.), and if you like to measure things accurately, use a noiseLevel Re to figure out how much more sampling you may need (about twice as much for half the noise level.).
There is no way to achieve that directly, but it’s easy enough to use a rendermask if you need that to happen: just render the second bit with fewer max AA subdivs.
This would make sense, perhaps, for a still only, however.
The texture may well need AA because, f.e., it would need continuity across frames.
Ultimately, the sampleRate Re will let you know where work was spent: if your texture is drawing too much sampling, it’s because the detail is much smaller than a pixel.
If it’s a still, then you can ignore this and use a rendermask with lower AA but if it’s an animation, then you will need it sampled “enough” for it to not sizzle across frames (any texture size at any point in time could have parts smaller than a rendered pixel…).
The user has no way to know when “enough” is (but one sure can want things different. hence the rendermask.).
Last but not least, texture sampling per se’ is trivial, compared to everything else, these days.
If it drives some other complex shader parameter, then sampling will likely be a must and, just as likely, will be drowned by the rest of the sampling the image requires.
All this gets further complicated by the fact that MIS comes into play (as a normal scene is lit with contrasty lighting), and so the sampling for the noise could be completed, but more sampling may be needed for the lighting to clear up and the lights may introduce it, regardless of the node being lit.
