Sampling tutorial for V-Ray 3.20

[^Lele^]

http://forums.chaosgroup.com/showthr...508#post666508

[Gilban]

Thank you everyone!.....Really usefull guys....

[Deflaminis]

Excellent thread here, very informative. Thank you all for taking the time to do this, it's been invaluable.

I wonder, dark scenes obviously are more noisy and in the past I have overexposed by 2-3 EVs to get a cleaner/faster render and saved out a 32bit image. Then in post reduced the exposure by 2-3 evs, setting it back to my original value. Is there a practical downside to doing things this way, other than leaning on the AA a bit more for bright areas?

Given that high contrast/overbright pixels take a bit longer to clean up, I've noticed that it seems it's still faster than underbright/too dark ones. My imagination?

[vlado]

I wonder, dark scenes obviously are more noisy and in the past I have overexposed by 2-3 EVs to get a cleaner/faster render and saved out a 32bit image. Then in post reduced the exposure by 2-3 evs, setting it back to my original value.

Hopefully this will no longer be needed with the upcoming service pack.

Best regards,
Vlado

[PIXELBOX_SRO]

so we started using the new sampling method and i have to say its acting weird.
Some of the scenes with the same settings take 3-5 hours over DR on 10 nodes (static interior scenes A4/300dpi format)
But others take easily 30-50 hours :-/
Thats a huge performance difference and one never know when the scene will take more or less given the same settgins.

We go with 1/24 for AA
noise threshold @ 0.005
MSR @ 50

[vlado]

so we started using the new sampling method and i have to say its acting weird.

Please don't post on the public forums about the nightly builds I did say it is still in development - the algorithm went through 100 revisions and it depends on which nightly you picked. If you have any issues please email me directly to vlado@chaosgroup.com

With that said, I've run the algorithm through several scenes to figure out how it behaves with respect to MSR; all data shows that MSR higher than maybe 16 is completely impractical and will actually increase your render times. MSR about 6 to 8 seems to perform best. If you are interested, I can email you that data as well.

Best regards,
Vlado

[PIXELBOX_SRO]

oh i am not talking about nightly...i am refering to your tutorial Vlado with controling the scene only settgin AA, noise threshold and MSR ....i would not dare otherwise )
cant be using nightlies in production ...woudl be too hard to troubleshoot

[vlado]

oh i am not talking about nightly...i am refering to your tutorial Vlado with controling the scene only settgin AA, noise threshold and MSR ....i would not dare otherwise )
cant be using nightlies in production ...woudl be too hard to troubleshoot

Ah ok In any case, MSR 50 is way too much. Lele wrote a small script for me to run a scene through different MSR levels and I've run a few scenes through it. In the majority of cases, values 6-8 were optimal or very close to the optimal - both higher and lower values increased render times, sometimes quite substantially.

Best regards,
Vlado

[Morne]

Vlado, I was very optimistic when you originally posted this. I tried it on some small test scenes and I was very happy with the render times and the quality of the render.

However, when I tried it on actual projects, the speed took a MAJOR HIT. The quality wasn't bad, just the speed sucked. For example I did an exterior house with 3D grass, trees etc, and I was getting render times of 7 hours on a single image of 3200X1800 which seem a bit extreme.

[vlado]

The quality wasn't bad, just the speed sucked. For example I did an exterior house with 3D grass, trees etc, and I was getting render times of 7 hours on a single image of 3200X1800 which seem a bit extreme.

That would be odd. If you can get me that scene it will be helpful. Those settings worked very well for me across a variety of scenes, including exteriors with tons of grass and trees, but there might be other things going on in your scene or you may have changed other settings too.

Best regards,
Vlado

[Morne]

That would be odd. If you can get me that scene it will be helpful. Those settings worked very well for me across a variety of scenes, including exteriors with tons of grass and trees, but there might be other things going on in your scene or you may have changed other settings too.

Best regards,
Vlado

Yes sure if we can keep it confidential. The Max file alone is 611MB, I'll zip it all together with maps etc and upload it somewhere. Can I mail you to vlado@chaosgroup.com with the download details?

[vlado]

Yes sure if we can keep it confidential.

Yes, of course.

Can I mail you to vlado@chaosgroup.com with the download details?

Yes, that would be perfect.

Best regards,
Vlado

[Morne]

Yes, of course.

Yes, that would be perfect.

Best regards,
Vlado

Confidential mail sent, together with a special request...

[Oleg_Budeanu]

I have a question about this workflow - how do i control GI quality ? For example i have a scene now and using this workflow comes out pretty good , though for some reason GI is a bit out on noise level and i would like to boost it , any ideas how to do that ?

[joconnell]

Are you looking at the normal, vrayglobalillumination element or the raw? Same as any other bit of secondary shading it'll be controlled by the msr or if you're not setting the subdivs multiplier to 0.0 to know out all the local subdivs, you can up the quality of your brute force gi directly to an amount larger than what the MSR would add.