So there is no correlation, everything is link to the AA only?
Here is an example (NO GI for now) : 1 HDRI as key light + one Vray plane as fill light. 100% Same setting as yours.
Send me the scene.
Not everything is linked to the AA though.
You have introduced completely different lighting…you will need completely different settings.
Just because the Sample rate pass looks correct, doesn’t mean the Beauty render will.
You can have perfectly optimised AA settings and completely dog shit subdivision settings. you need to balance BOTH
So what will fix this scene?
Go to the dome light and increase the subdivisions until the Lighting pass is more or less noise free.
Nope, not homework for me to do! I’ve been threatening a video for a while and I’ve the time to get it sorted now. Looking forward to sitting down with your materials too - some cracking results here.
It looks to me like ur Dome lights its night or sunset. Which are the hardests IBL cases to clean up. I suggest 128 Sub D on dome and light. Also use Adaptive subdivision with values 1-2. It should render a lot faster than Adaptive DMC and be quite clean. If its still noisy Id up the IBL subD.
Another poor soul lost to the misguided documentation.
The Adapative DMC sampler is best in ALL cases when utilised correctly as you have the most control over where you sample the image.
Don’t worry, I used to only use adaptive subdivision for years as it was the only way I knew how to get rid of noise.
Well its debatable.
I get smooth light with 12 samples on Adaptive Subdivision where as in DMC I need 128 samples. It also renders a bit longer in DMC. There is nothing else in game here just plain sampels and the way AA look at pixel.
Im just saying. Each scene and set of problems is better solved using either Subdivision or DMC.
Do you mean the min / max value is 1-2?
That is crazy. Remember, the Max value divides the global samples. So having 300 subdivisions isn’t the “actual” number. It’s relevant to the max AA samples.
Indeed, it is! I’m wondering if cranking up the intensity multiplier of HDRI will not get it render faster? Sadly, it can’t be apply in this scene.
But it seems that lighter is the scene and faster is the render.
I thought you love challenges 
Well like I said its very difficult one. If you are going to use Adaptive DMC - Blurr HDR a little bit. Depending on the light sources in image it will depends. You will loose a little shadow accuracy but that should be fine with night shot. Remember the smaller the light source the harder it will be to render for vray. So blur should give it a little speed up.
Using Adaptive DMC =
2/32 min max.
IBL 256-512 samples.
That should get u started on LIGHT ONLY Pass. I don’t discuss reflection that’s way more complicated.
However I would suggest using Adaptive Subdivision. In which case you are looking at:
1/2 min max samples AA
64-128 samples on IBL. That should render as long as adaptive DMC but it will be cleaner. Shader samples depending on gloss if its around 0.7 I suggest 24… Its hard to just shot out the numbers…
Hope this helps…
I think that may be because you’ve got 22,500 light samples (300x300 = 90,000 samples / 4 AA samples). Is that correct?
Yes, Light subdivision is 300 and AA is 1/4.
Will try to push the light again, but the render time is really going up too.
→ no , doesn’t make any improvement even with light subd set to 600.
Email the scene to me at
admin@sathe.com.au
This may be one of those rare cases where you need to adjust the noise threshold (Not the same as CLR threshold)
I’m also up for testing Adaptive Subdivision again and potentially starting a shit storm haha
AHahah I’m almost certain it’s not going to be of any benefit whatsoever and there is simply something in the scene I’m missing.
(I’ve had to balance HDR’s countless times now so I’m well aware of the noise dome lights bring to the table :P)
But if I am wrong, I’ll happily fall on my sword 
Hey fraggle can you post the HDRi?
I have done a few tests with some of mine, and no real issues. Just one or 2 but not noise problems, more object edge and shadow edge.
I think reducing the Clr thresh should fix it or playing with the HDRi resolution. Sometimes I wish DMC had Object outline like subdive.
I will post images shortly.
Guys Using AA to clean up light noise samples is the worst possible thing you can do.
This thread is an absolute mess.
Considering breaking it into Support and discussion but then we lose the potential of overtaking the biggest thread on the chaos group forum haha.