Cleanup noise in Elements: Reflection and Shadow - Also LOOONG render times

[Morne]

Hi all

I have a LWF scene that is taking very long to render and is very noisy at 720p. I added a whole bunch of RenderElements to get to the bottom of it. So almost all the elements are nice and smooth Except for the VRayReflection and VRayShadow elements. The VRayReflection Element is VERY noisy. Whats' the best route to clean it up? Simply add more subdivs in the reflection part of the materials? Or is there another way with maybe dmc and dmc sampler to do it?

My current settings:
reflection subdivs on materials vary between 16 and 64 (which is already high)
Scene lit with only 3 cove lights with VRayLightMTL multiplier 3, compensate camera exposure ticked, direct illum ON and 128 subdivs

Adaptice DMC: 2, 32, 0.005 thresh UNTICKED
IRMap: Medium Preset with HS subdivs 50 and Inter samples 40. Detail Enhancement on and Check sample visibility ticked
Light cache 1000, 0,02, prefilter ticked on 10, interp samples 10, use light cache for glossy rays TICKED
DMC Sampler: adaptive 1 thershold 0.005

It takes 4 X i7 pc's 2 hours to render 1 frame with DR at 720p!!!

Any hints?

GI Element has very minor minute noise hardly visible

EDIT: Max 2012 scene file dmc.zip attached a bit further down in this thread at post no 8
EDIT2: Max 2011 scene file just 2 posts below that at post no 10

[AJ Jefferies]

Hi Morne,

Would it be possible to save out a stripped down scene file with maybe one or two problem/noisy obects so I could take a look?

Cheers!

AJ

[Morne]

it's most of the objects as all objects have glossy reflection.
Main one however is default VRayMTL with diffuse 200,200,200
white reflection fresnel ticked
glossy 0.85
64 subdivs

[Morne]

ok so I've switched from IRMap to Brute force and put subdivs to 256 and also on my material reflection put 256 subdivs for reflection and glossy 0.625. For DMC changed to 2,6
NO noticable change!
Anynody have any thoughts?


This is is zoomed to 200%
You can also notice some noise on my metal shopfront frames

[zeronove]

afaik, if you need 32 as max for adaptive dmc you have to raise mat & lights subd a lot more then 64 or 128 (I'll say 256 and raise clr threshold to 0,01 or more) because right now you are leaving all the job to be done by image sampler (that in 2.0, for me, is slower then in 1.5)
hope it helps

[Morbid Angel]

while gi has minor noise, it contributes to render time, detail enhancement is brute force with a radius. Check this render out. Its all brute force/lc with glossy reflections/refractions. All in all render time is 2.06 on my 12 core xeon.

[Morne]

Check this render out. Its all brute force/lc with glossy reflections/refractions. All in all render time is 2.06 on my 12 core xeon.

At the size posted it took 2 hours? Or is the posted version smaller?

[Morne]

ok here is a simplified version of my scene above. After 40 min of render, this is all that is rendered, its estimated another 3 hours to go. And it has LOTS OF NOISE!
What can be done to clean up the noise and also speed up render times with brute force?


And also,
here is the max file:
dmc.zip

It's a max 2012 file with VRay 2 latest stable public version.
This simplified scene has most of the major materials from current bigger scene.

[zeronove]

ehm, I don't have 2012 installed here, could you post a 2011?

[Morne]

ehm, I don't have 2012 installed here, could you post a 2011?

sure, here you go
dmc2011.zip

[Morne]

hmmm ok, so I've changed to following:
in the main material in options I changed cutoff from 0.01 to 0.001 and also I've set "treat glossy rays as GI" to ALWAYS instead of "only for GI rays"

and now its rendering about 75% FASTER. It's still dog slow I think. Took 1 hour 29 minutes on my machine.

ALSO

I tried switching to reinhard with burn of 0.5 and "adaptation only" unticked instead of full LWF and also it rendered FASTER with a time of 54 minutes total. Also for reinhard I needed to bring the exposure down with a smaller value to get a nice range (-0.7) and with linear I needed to bring it down by as much as -3.6!

So I'm geussing then if I just expose properly using linear, theoretically I should get same render times as with reinhard.

I would like to however clean the image up some more AND bring the render times down even further.
Any thoughts?

[Morbid Angel]

I had a quick look at your scene. I would not recommend using direct lighting from vray light material, it will only make the image more noisy. You also have amb occlusion checked, it will make it also slower and look wrong in terms of gi.

[Morne]

I had a quick look at your scene. I would not recommend using direct lighting from vray light material, it will only make the image more noisy. You also have amb occlusion checked, it will make it also slower and look wrong in terms of gi.

ok then would you say use vray light in "mesh" mode and select my cove lighting that currently has the vraylightmtl on it?
would u then do a seperate dirt pass for the AO or how?

[Morbid Angel]

My understanding is that its a clean interior, why would you need a dirt pass? Dirt/ao gives unnatural darkening in areas you normally would not have.
For the lighting i would not use mesh light either. Mesh lights produce more grain then regular lights. In this case, you are better off just using gi in that area from light mtl.
In terms of render time, any time you render with bruteforce expect your time to be high.
For fast render time you got irradience map (obviously not as good).
I will show you a diff setup when im home later today.

[zeronove]

playing with your file I'm just going in the same direction Dmitry suggested

till now, IMHO the main grain generators are:

- vrlightmtl with direct checked: do you really need it? ...from my test I'll say a big no, go with GI only
- ao pass: that one it's easier, just raise subds a bit

IMHO I'll switch to IR/LC + AO (with extra text could be better); it's a fake but it's faster; at the end the final decision will be based on the real complexity of the scene (if your renderfarm could afford long render time then go with brute forcr, otherwise stay with IR/LC