"Devide shading subidvs" leaves Vray working linearly? Is this correct?

[joconnell]

Righty - The coupling of subdivs that happens in vray is definitely as simple as just the ui dividing all your subdivs by the max AA - that's all it does, there's nothing going on at a core level. Beyond that all the sampling is just vray dealing with noise.

On image A versus image B I totally understand and really what's happening here is just that there are many different ways to try and achieve a clean image, and it's probably possible to get a clean image using any of those methods but some methods are better suited to certain scenes than others. The various approaches to sampling really boil down to do I want to spend time optimizing my settings for quicker renders, or do I want to use less controls and have sightly slower renders. Vlado would love to boil it down to just one method of working as it means the ui could be simplified, learning vray could be simplified and we could all move on - the problem is there's people like me who endlessly fiddle with vray and we've gotten used to a different way of working from universal type setups. It's always really tough to remove a bit of functionality from software, even if it's a worse way of doing things as you'll always find someone who relies on it heavily. The autodesk guys for example would love to throw out editable mesh and editable poly as it causes a tonne of problems for them, but so many people use max in such different ways that there'll be someone that'll cause uproar if their favourite button gets removed!

In terms of the extra check box, what it did was add in yet another method to optimize vray and while that works really well for some people, it's adding in another control and Vlado would really love to simplify, not add to the ui. The problem it was put in to resolve, it's definitely solved (which is only to stop max AA from dividing all your other ui controls) but it might be that you're expecting it to do something different?

[RockinAkin]

Recon - My DMC Calculator might be helpful for you see exactly what the 'Divide Shading Subdivs' checkbox does to V-Ray under the hood:

http://www.cggallery.com/tutorials/dmc_calculator/

As John said - all it does is stop the AA max samples from dividing your scene's secondary samples values (lights, gi, materials, etc).
The calculator will show you how the formula changes when you check it on and off.

[LudvikKoutny]

Recon - My DMC Calculator might be helpful for you see exactly what the 'Divide Shading Subdivs' checkbox does to V-Ray under the hood:

http://www.cggallery.com/tutorials/dmc_calculator/

As John said - all it does is stop the AA max samples from dividing your scene's secondary samples values (lights, gi, materials, etc).
The calculator will show you how the formula changes when you check it on and off.

Seriously... i know what it does. I know exactly what it does. I fully realize what it does and at the very same time i realize how it limits me and how it makes optimization process harder.

In any other renderer, if you have noise, you increase sampling of something. In Vray, that may sometimes result in more noise. If you increase AA for example. So it's harder to optimize.

Or if you have clean image, and want to optimize rendertime, you remove samples from some material or light. Again, in Vray, that often leads to worse quality, but no change in rendertime.

In any conventional renderer, it's simply about tweaking the performance/quality ratio in a linear way. That way, it is very easily predictable, very fast, and quite efficient process.

In case of Vray, it is searching for an exact sweet spot between too little samples, resulting a lot lower quality than what you optimally should achieve in given rendertime, and too many samples, resulting in excessive rendertime with no growth of quality.

These sweet spots are lot less predictable, take lot longer to find, and are lot harder to balance. This makes any optimization attempt less beneficial, if at all.

So as i said before. There was never a time, where i would get too bad rendertimes in Vray, for given quality. But there was also never a time, where i could get rendertime which would make me say "Whoa, that's really fast".

[Dariusz Makowski (Dadal)]

Have you tried universal sampling? I never really trusted it until recently and now I'm quite surprised how good/easy it is. Also I only have to control 1 parameter to get it to work so wow.

[LudvikKoutny]

Have you tried universal sampling? I never really trusted it until recently and now I'm quite surprised how good/easy it is. Also I only have to control 1 parameter to get it to work so wow.

Yes, i did And it works. But it's that exact case. You never get really bad rendertime, but you never get really good rendertime either

[Dariusz Makowski (Dadal)]

HUmh, tricky. I just started testing it will need to do more tests before I can whine about it hehe

[joconnell]

Seriously... i know what it does. I know exactly what it does. I fully realize what it does and at the very same time i realize how it limits me and how it makes optimization process harder.

Sure I getcha - as you mention increasing aa then that's the only killer really, adaptiveness can be a bit of a pain in terms of turning down material or light samples upping AA time but you'd be getting utterly destroyed by render times if you didn't have it. A few friends are using another renderer which is nothing but monte carlo raytracing so everything is grainy. It's nice looking grain admittedly but if they want to go to a clean render, they can't as times just kind of explode. That's linear control but it still means a lot of penalties.

[LudvikKoutny]

Sure I getcha - as you mention increasing aa then that's the only killer really, adaptiveness can be a bit of a pain in terms of turning down material or light samples upping AA time but you'd be getting utterly destroyed by render times if you didn't have it. A few friends are using another renderer which is nothing but monte carlo raytracing so everything is grainy. It's nice looking grain admittedly but if they want to go to a clean render, they can't as times just kind of explode. That's linear control but it still means a lot of penalties.

Linear control does not mean monte carlo path tracer. Those are completely unrelated things. Arnold having linear sampling scheme is not a result of it being a path tracer. Mental Ray for one isn't... Modo renderer has linear sampling and it isn't MC path tracer either. Same with finalRender. All of these mentioned renderers are admittedly worse than Vray, but that's not result of Vray having subdivision based sampling scheme.

[joconnell]

Sure of course, just giving an example of a renderer that is linear and it ain't exactly a bed of roses

[Lupaz]

It's nice looking grain admittedly but if they want to go to a clean render, they can't as times just kind of explode.

They can't yet. It's a matter of time.

[joconnell]

Absolutely! Given time everything will improve and while I've invested a lot of time into learning the software I use now if something else comes along that's clearly better then it's time to re-evaluate. The thing is to pick what's best for what I have to do right now!

[Lupaz]

True. It was kind of stupid of me. But what I meant is that may be it is time to give V-ray some ways of using it that are more...hmm... linear.
Just for those cases when you need to analyze the scene and compare with what you where doing before, or even just so we don't even think of the possibility of going to another renderer in the future. Just my 2 cents.

[joconnell]

Sure. What might be the handiest thing to do in that regard then is go universal a la vlado, let the AA sampler be the only one doing all the work and then just use your noise threshold as the single control between speed and quality.

[Lupaz]

Wouldn't that be the opposite of what we are talking here?
Some say they love the universal settings (which is contrary to what it was written for so many years by Vlado himself, that there are no universal settings).
Others say it's a completely wrong approach.

I believe that other render engines are noisy because they let the secondary rays to do their job, as opposed to the AA. Isn't that right? Vray would be something closer to this with "divide subdivs" un checked. Am I way off here?

[alexyork]

Sure. What might be the handiest thing to do in that regard then is go universal a la vlado, let the AA sampler be the only one doing all the work and then just use your noise threshold as the single control between speed and quality.

I'm increasingly working like this. For me it makes far more sense to spend time working on the scene itself and then have a noise control for render times. During testing I'll render low-res drafts with a reasonable noise level. For finals I'll just let it cook longer at full res. It's a nice workflow. More like what Maxwell is doing.