Understanding DMC Sampler

[grantwarwick]

It's still quite a tricky subject.
Especially when DOF is involved. As That seems to be a bitch to tackle no matter how you look at it.

Keep the color threshold low to begin with- .01 Then raise it last. That's what I'd do.

[3LP]

Hey mates,

I tried this whole theory here in a new interior,
And I don't know what is going on, but my render times are going through the roof...

Here is my render with passes :

picture uploader
Render Settings:


Rendertime : 12 hours on a i7-3930k 12 threads 4Ghz.

materials are all going from 128 to 256 subdiv.
I reverted back to IR with detail, and light plane with light portal because BF and dome light just killed it, a 16*16 bucked stayed there for 30 minutes and resolved with loads of noise...

I don't know what's wrong with this scene,
does anybody has any thoughts?

Thanks
Stan

[Dariusz Makowski (Dadal)]

Yea wow, a lot of noise. Can you maybe send me scene?

Thanks, bye.

[3LP]

Hey,
Thanks DADAL, I just send you a PM with the file.
Grant, do you want to have a look as well? :-p

thanks guys!

Stan

[peterguthrie]

Tried a new test over the weekend:



'New method' was with the following settings:

DMC Image Sampler

min 2, max 32

clr threshold 0.015

Settings: DMC

Adaptive amount 0.85
Noise threshold 0.01

Lights

generally 80 (use VMC script)
vraydomelight 240

GI

240 subdivs

FOG (if applicable):

160 subdivs

Materials

generally 80 subdivs (use VMC script to set)
other glossy mats than have 0.5 - 0.75 glossiness: 160
other problematic materials: 240
plasterboard (as noise always shows up in it): 320

Camera

Through trial and error I found 24 subdivs to be the fastest

Sure there are slight differences between the two, but I can live with that

[3LP]

Nice,
Just 3 times the speed, yeah there are some slight differences, I wonder what's the "right" one... did you aimed to have your materials like the universal or like the "new method"? (I like the new method more)

This tend to work really well I think for exteriors, but I,m not that sure when it comes to interior where you've got lots of GI light...
Even your scene that you posted before has a very wide opening in the ceiling, so I don't know if it can be considered as a typical interior scene.
Keep it up, I would like to see more example with this method on all type of scenes
(Although my attempt pretty much failed and I'm not sure why or how to fix it if I don't delete the HDRI)

Stan

[peterguthrie]

Nice,
Just 3 times the speed, yeah there are some slight differences, I wonder what's the "right" one... did you aimed to have your materials like the universal or like the "new method"? (I like the new method more)

This tend to work really well I think for exteriors, but I,m not that sure when it comes to interior where you've got lots of GI light...
Even your scene that you posted before has a very wide opening in the ceiling, so I don't know if it can be considered as a typical interior scene.
Keep it up, I would like to see more example with this method on all type of scenes
(Although my attempt pretty much failed and I'm not sure why or how to fix it if I don't delete the HDRI)

Stan

probably the universal one is more correct..

how about I try it on this scene http://www.flickr.com/photos/pg/8536...57632942671928 it was a killer to render

[vlado]

how about I try it on this scene http://www.flickr.com/photos/pg/8536...57632942671928 it was a killer to render

Can you get me this scene? I'm running V-Ray through various scenes to try and profile the performance; it is useful to see which parts of the code take the most time so that they can be optimized.

Best regards,
Vlado

[joconnell]

Stan there's a few problems with your scene and to be honest there's no magic in grant's numbers aside from them being good for scenes that have a lot of advanced lighting and materials. You'll have to tweak and balance each scene every time to get decent results, a preset won't do it unfortunately :/

[peterguthrie]

Can you get me this scene? I'm running V-Ray through various scenes to try and profile the performance; it is useful to see which parts of the code take the most time so that they can be optimized.

Best regards,
Vlado

Of course Vlado, give me a bit of time to tidy it up 1st though

[3LP]

I understand that, but still 12h for a 700x900 image on a 12 thread i7 with still noise...
maybe that's what it takes, but it seams so excessive, no?

Vlado, do you think you could have the same check on my scene ?

Thanks

Stan

[joconnell]

Yeah that's nuts alright. I'd bring down your AA to half - try 1 / 16, I'd up the samples on the light outside the room as that's very very noisy and up the subdivs on the reflections on your wall and floor - maybe double them. I'd do an override material of a grey to make sure your lights are clean first though, as you can have noisy reflections, or materials reflecting noisy lighting.

[Morne]

Yeah that's nuts alright. I'd bring down your AA to half - try 1 / 16, I'd up the samples on the light outside the room as that's very very noisy and up the subdivs on the reflections on your wall and floor - maybe double them. I'd do an override material of a grey to make sure your lights are clean first though, as you can have noisy reflections, or materials reflecting noisy lighting.

John, how's your video coming along that you mentioned once or twice on the subject?

[Dariusz Makowski (Dadal)]

3LP - I had a play with your scene and so far its the HDR that is very challenging for sampler to clean up. I think with such extreme lighting you might be better off with sun and sky node. At the moment I run at 400 samples and its still very noisy in diffuse pass so... lots of stuff to optimize.

[peterguthrie]

Is it one of my hdri skies? Have you tried reducing the gamma to 0.75 to get more direct light in there?