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

[Lupaz]

Hi all,
I am aware that this was probably discussed in the thread "mastering vray...", but it would take me too long to read over 1000 posts.


I'm not sure if I understand correctly how vray works with and without "divide shading subdivs".

As far as I understand, because computers are faster now, some people would like to increase the realism of renderings, while keeping the renderings smooth, by adding more samples to the shading rays as opposed to samples in the image sampler. That would mean more accuracy in shadows, scattering, etc. no?

With Vray as it was before V3, this was very hard to achieve because vray was changing some parameters automatically (by dividing shading subdivs) as we were adding more samples to shading rays. Correct?

By un-checking this option, the calculations become linear (sort of?), which would be nice and I understand why some users want this.

However, un-checking this, makes renderings way too slow. Does this have to do with that the default subdiv values for materials, lights, etc, are too high if they aren't divided?

Thanks!

[vlado]

As far as I understand, because computers are faster now, some people would like to increase the realism of renderings, while keeping the renderings smooth, by adding more samples to the shading rays as opposed to samples in the image sampler.

It means that someone whined a lot on the forum that other render engines worked with this option disabled and were therefore so much easier to optimize. So I added the option as a test, and as expected, they found out that it was not really such a big plus anyways.

My own preferred method for adding more shading samples is through the "min. shading rate" parameter.

The default values for materials/lights/etc are indeed way too high by default, so you'll need to readjust everything.

Best regards,
Vlado

[RockinAkin]

...some people would like to increase the realism of renderings, while keeping the renderings smooth, by adding more samples to the shading rays as opposed to samples in the image sampler. That would mean more accuracy in shadows, scattering, etc. no?

Yes, in some cases allocating more samples to the shading subdivs means more accurate reflections / shadows / etc.

With Vray as it was before V3, this was very hard to achieve because vray was changing some parameters automatically (by dividing shading subdivs) as we were adding more samples to shading rays. Correct?

Close - as we were adding more samples to the IMAGE SAMPLER, v-ray is reducing shading samples to compensate under the hood. This meant if you raised your Image Sampler settings, you'd have to once again raise your shading subdivs throughout the scene to get the same amount of shading samples taken as before.

By un-checking this option, the calculations become linear (sort of?), which would be nice and I understand why some users want this.

Yes, basically - what you set is what you get. Increasing the image sampler's settings no longer effects your shading subdivs when 'divide shading samples' is unchecked.

However, un-checking this, makes renderings way too slow. Does this have to do with that the default subdiv values for materials, lights, etc, are too high if they aren't divided?

Right. If you're going to go into full-control mode and disable the 'divide shading subdivs' checkbox, you'll probably have to go through your scene and reduce the subdivs value of all your lights, gi, materials, etc, to a lower value than the default of 8 subdivs.


All that said - I do think Vlado is right that simply using the Min Shading Rate is the best way to quickly balance your scene's Image and Shading samples.
Having the option to fully control v-ray's sampling by unchecking 'divide shading subdivs' is a great feature - but not needed most of the time.

[Lupaz]

I do think Vlado is right that simply using the Min Shading Rate is the best way to quickly balance your scene's Image and Shading samples

Got it. Thanks to both.

But now I'm confused:
I always used Global subdivs multiplier with that purpose, since it doesn't affect AA, according to the manual.

Would you guys say that it's a good idea to use both? Meaning, increasing (or decreasing) the global subdiv multip, AND playing with "Min shading subdivs?

And if I use "global subdiv mult", how is it affected by the "v-ray is reducing shading samples to compensate under the hood" when "divide shading subidvs" is checked?

[vlado]

Would you guys say that it's a good idea to use both?

No; would be a lot less confusing it you stick to just one of them.

And if I use "global subdiv mult", how is it affected by the "v-ray is reducing shading samples to compensate under the hood" when "divide shading subidvs" is checked?

V-Ray first multiplies the samples, and then divides by the AA subdivs, same as before.

Best regards,
Vlado

[LudvikKoutny]

Actually, i tried to take advantage of it several times, but i have a very strong suspicion it still does not work completely right. Even when divide shading subdivs was OFF, I often ran in a scenario where increasing shading subdivs of for example DomeLight, increased quality and decreased rendertime. Such thing would never ever happen in for example Mental Ray.

Although i am not sure if that's an actual bug, or just inability to completely separate AA sampling from shading due to the Vray core limitation.

So in the end, it may not be useless feature, but it may just be implemented wrong.

[vlado]

Actually, i tried to take advantage of it several times, but i have a very strong suspicion it still does not work completely right. Even when divide shading subdivs was OFF, I often ran in a scenario where increasing shading subdivs of for example DomeLight, increased quality and decreased rendertime. Such thing would never ever happen in for example Mental Ray.

Of course the same thing happens with mental ray and I can prove it.

It is implemented correctly.

Best regards,
Vlado

[LudvikKoutny]

Okay, here is my proof:

I have set up a scene in both Vray and MR in as identically as possible.

There are both scenes as well as test render results attached in the rar file below. Settings and rendertimes are written in the file name, in a following format: rendererName_qualityValue_rendertime(minutes-seconds)

For Vray, i increased DomeLight subdivs, for MentalRay, i increased IBL quality setting.

Results are obvious. In MentalRay, rendertime grows linearly with increased quality setting, and noise amount reflects that. In Vray, increasing DomeLight subdivs with "divide shading subdivs" disabled, results in both faster renders, and cleaner output. Up to some point, where you start to waste samples. While it's an improvement, compared to old mechanism, it still isn't reliable enough to be intuitive beneficial.

[grasshopper]

Could it be that you are using an adaptive image sampler and that the increased Domelight subdivs are allowing it to work more efficiently?

[LudvikKoutny]

Could it be that you are using an adaptive image sampler and that the increased Domelight subdivs are allowing it to work more efficiently?

Actually it's exactly that, but the point is that if you decouple AA from shading, it should not happen.

[vlado]

Okay, here is my proof:

<shrug> Here are my tests with mr; will get to the V-Ray tests a bit later. You can see that increasing the IBL quality initially results in worse noise and faster render times, which is quite similar to what happens with V-Ray. After a certain point, increasing the IBL quality gradually improves the noise and increases the render time. When the render time gets to the same level as the first image, the noise in some parts is better (but worse in others - I think it depends on the HDR image).

It's a bit difficult to figure out what happens with mental ray, as it doesn't allow to enter samples for IBL directly - I'm not quite sure how the "quality" maps to the number of samples. I'm also not sure why the unified sampling has both "quality" and contrast/noise thresholds and how they relate to one another (as they both have an effect on the image).

Best regards,
Vlado

[LudvikKoutny]

That's strange. I am not sure 0 is even valid value for quality, so it is hard to tell what's going on when the quality parameter is set to 0.

As for other results, i do not have that particular HDRI map to verify, but i struggle to believe that increasing IBL quality from 0.2 to 0.3 introduced so much more noise...

EDIT: Oh, my mistake, i did not attach the environment map. Find it below:

Dumb mistake, really... sorry

EDIT2: Quality in MR's sampler is sort of like global subdivs multiplier in Vray's DMC settings. Contrast thresholds are sort of like DMC sampler threshold, but do not have THAT much of an impact, as AA is properly decoupled from shading.

[vlado]

Here are also some tests with the V-Ray scene; however I can't replicate your results - increasing the dome light subdivs always leads to less noise and longer render times. Maybe it's due to the particlar HDR image; not sure.

Best regards,
Vlado

[LudvikKoutny]

Ok, i re-tested MentalRay scene with some higher contrast HDRI instead of 8bit jpeg environment map i used before. And getting same consistent results. Higher quality parameter always results in less noise and longer rendertime. You must have mixed up your results somehow.

I am attaching the scene, my results, as well as high contrast HDRI i used. The HDRI is courtesy of Jeff Patton. He released it in one of his free to download HDRI packs, so i hope he won't mind me posting one of them here

https://dl.dropboxusercontent.com/u/...trast_HDRI.zip

I will try your Vray scene now.

EDIT: Oh, it's not a scene, just results... nevermind then

[LudvikKoutny]

Ok, so with contrasty HDRI map, i got same results with MentalRay... means with increased sampling, i observed improvement in quality and growth of rendertime.

Now, i tried Vray with same high contrast HDRI, and i observe the same as before. With increased sampling, i observe improvement in quality and decline in rendertime, up to some point, where rendertime starts to grow again due to the wasting of samples.

I am quite confident when you did your tests with Mental Ray, you mixed up your results somehow. Otherwise i can not explain it. I am attaching the Vray version of scene with high contrast HDRI, and HDRI map used in this scene can be found in the Mental Ray version of the scene in the post above