V-Ray Render Optimization - an in-depth Guide (call for Before/Afters)

[RockinAkin]

Can you explain a bit more what exactly it's changing in the subdivision?

Min Shade Rate basically raises the maximum number of secondary samples taken in your scene, regardless of what your AA is set to.

You can see exactly what's happening with my DMC Calculator

[bennyboy]

The "Adaptive subdivision" sampler is the odd one out, as it does not divide the shading subdivs by the AA samples.

Been thinking more about "Adaptive Subdivision", which tends to be the fastest solution for me most of the time...so, Vlado, this means, since it's not dividing the shading subdivisions, the shading subdivisions (either through the "Min Shading Rate" or individually through the materials, lights, etc.) can be much, much lower to have the same effect as the Adaptive method (as shown in Rockin's tutorial)?

[vlado]

Been thinking more about "Adaptive Subdivision", which tends to be the fastest solution for me most of the time...so, Vlado, this means, since it's not dividing the shading subdivisions, the shading subdivisions (either through the "Min Shading Rate" or individually through the materials, lights, etc.) can be much, much lower to have the same effect as the Adaptive method (as shown in Rockin's tutorial)?

Possibly, but not definitely... With the adaptive AA sampler, you can count on it to "fix" any noise due to low subdivs for materials or lights. The Adaptive subdivision sampler is not very good at this - it assumes that the effects have high enough subdivs to appear smooth in the first place. Sorry for the vague answer; it's best to test for the particular scenes that you have.

Best regards,
Vlado

[bennyboy]

Thanks Vlado.
I think I know what you're getting at. Also, I realize now, I really didn't know the fundamental difference between Adaptive Subdivision and Adaptive DMC. Along that line, I will say that, though Rockin's tutorial is incredibly eye opening and just truly great, it didn't mention (I don't think) any of the other AA methods, particularly Adaptive Subdivision. He, however, recommends another tutorial that I also read and it explained Adaptive subdivision in-depth: http://www.interstation3d.com/tutori...yfing_dmc.html

[RockinAkin]

Thanks Vlado.
I think I know what you're getting at. Also, I realize now, I really didn't know the fundamental difference between Adaptive Subdivision and Adaptive DMC. Along that line, I will say that, though Rockin's tutorial is incredibly eye opening and just truly great, it didn't mention (I don't think) any of the other AA methods, particularly Adaptive Subdivision. He, however, recommends another tutorial that I also read and it explained Adaptive subdivision in-depth: http://www.interstation3d.com/tutori...yfing_dmc.html

I think (and maybe Vlado can correct me if I'm wrong) but the Adaptive Subdivision sampler is kind of depreciated these days. I never use it anymore personally, and I don't know of any studios using it.

Adaptive DMC, Fixed Rate, and the new Progressive sampler are all basically variants of the same sampler... so my tutorial applies to all of them. Fixed Rate is basically Adaptive DMC with the Min and Max subdiv values set to the same value, and Progressive is just Adaptive DMC rendering the entire image at once instead of bucket by bucket.

In fact, in my ultra-minimalist mindset, I wish V-Ray would simplify the UI by having having the dropdown to select samplers only have Adaptive DMC and Adaptive Subdivision, and then roll in the options of Fixed Rate and Progressive into the Adaptive DMC sampler's rollout.

[jstrob]

Very well explained! Thanks for this.

Only thing I would add is to suggest the use of the global subdiv mult instead of changing every subdiv individually. And then change only the individual subdiv that really need it. Most of the time I only need to tweak the global one.

Most of the time I will set the adaptative to min 2 and max 16 and then set the global between 16 and 32 and play with my noise threshold until I get a good balance. I always do animation so I use mitchel-netravali as an image filter. I also always increase lightcache subdiv to 3000. All other subdiv most of the time to 8.

I did that without really knowing what vray was doing under the hood but now i know why I do it and I'm better prepared to know what to tweak next time I get a difficult scene to render.

[fewlo]

Hi,

Hi, I've been using this method and have a nice clean blue sample rate pass, but still have noise from my lights, does this mean I've gone too high with my AA max? I'm at 1 min 18 max. The noise is apparent in my animation, it almost moves across the screen...

http://youtu.be/2LqvhgN6ihI

Once again, I must be missing something??

[RockinAkin]

What are your lights subdvision value set to?
What is your GI method?
Can you show the SampleRate pass?

It's a little hard to see whats wrong with Youtube's compression - are you talking about the noise in the upper middle walls that seems to stay stationary?

[fewlo]

yes that is the noise, and that is a good description, it seems to stay stationary.

light subdivs were 256, GI was Irr map & light cache, the thing that is throwing me is the sample rate pass, here's one of the frames:



If I understand the theory right, this is telling me that my AA is set pretty good, its just having to work hard for the trees outside of the windows. It is also telling me that my lights are clean as there's no red in those areas. I was happy with this and my beauty pass looked Ok as a still, since I've done post to my animation I've realised that the lights are very noisy, but I really thought this would be picked up in the sample pass? I upped the offending light to 1000 subdivs (!) which has really cleaned up that area but render times aren't great, however I still feel like I'm getting something wrong?

[RockinAkin]

How are you using Irradiance Map in this scene? Precomputed along the camera path, saved, and then loading from file?

Does the noise go away if you switch your primary GI method to Brute Force?

[Neilg]

the irradiance map doesn't make noise like that. those settings are crazy. what's the dmc noise threshold, are you using gamma correctly, what does the reflection, raw light & specular pass look like, is the noise appearing when you lift the levels in post (and how much do you lift the image in post), are you using a dirt pass. etc etc.

[Morbid Angel]

To me the noise looks like the adaptation is too high in those areas.

[vlado]

It would be best if you can get us a scene to look at; or you can send it to me directly to vlado@chaosgroup.com; otherwise it's just guesswork.

Best regards,
Vlado

[fewlo]

thanks for the replies, vlado I will look at getting a scene across to you once I'm back in the office Monday, although I don't think its vray playing up but simply my lack of understanding!!

dmc noise threshold is default, yes using gamma correctly, the noise is definitely from the lighting pass, I was using this pass as an 'add' layer in post so therefore making the problem worse, I'd not really noticed it in my test stills until I output my animation.

I'd arrived at using 1/18 for my AA with this threads theory, this seemed high but I ran with it as my sample pass was nice and blue, so maybe I've missed something as I'm unclear what 'adaptation is too high in those areas' means? At 1/18 I was comfortable with my lights at 128 or 256 subdivs, render times decent for my setup, but the noise is apparent in my animation!

As mentioned, to clean up this issue I upped to 1000 subdivs for the offending light which seems ridiculous, render times gone up too. The light is a simple vray sphere representing a bulb inside a typical lampshade, which is a solid material.

It amazes me how little I know about this stuff, when I think i've made a breakthrough I hit another problem! I really appreciate your time and thoughts on this.

[Neilg]

the noise is definitely from the lighting pass, I was using this pass as an 'add' layer in post so therefore making the problem worse,

You need to turn your lights up. sounds like they're too dim for vray to know that it needs to sample it properly. if you're rendering black on black, vray wont sample because the result will not be visible. then if you increase the brightness in post it will look awful because it's unsampled.
if you do a test render and artificially lift something like lighting or reflection a huge amount because it's barely visible, you should try to go into the light/material and match it to your intent so you can turn that off.

A good practice is to render a still early on, take it into photoshop with channels and adjustments to spend some time making it look how you want - then go back into max and copy the changes back to the materials/lights.