CHALLENGE - Show your skills (scene included)

[^Lele^]

mienda,
did you change something in the leaves shaders (both trees and bushes)
they looks more green.

i believe that's just removing the floating poly sheet on the outside of the windows.

[cecofuli]

I Lele,

the goal is to reduce the rendering time with a good noise with BF+LC with sub-pixel OFF
I think by reducing the shutter speed, leaves look more green, like in my next test:


(*) remove double window glass
(*) reduce shutter speed to 250
(*) min RGB level for dark shader color as 3
(*) Min glossy level for any shaders : 0.5
(*) Ceiling lights as default (no one big light) as client request.

(*) Sklight portal : simple (greatly reduces the time rendering)
(*) AA 2 16
(*) VRayAnczos filter 2.0 (good for smooth the noise and the bright AA aliasing pixel in the ceiling)
(*) BF 48
(*) LC 2500 screen 0.01 - Pre-filter 10 - Retrace ON
(*) DMC default

(*) All MTL subdivs = 8 except 4 shader: dark wall = 128 | white ceiling = 128 | metal windows frame = 64 | Dark glossy glass = 64
(*) Light subdivs 12
(*) DOF = 6 subdivs

The rendering time now is better (1h 52min with I7 970 @ 4Ghz) and the noise acceptable.
I don't know if it will be possible reduce more. The image is a PNG.

[mienda]

Hi cecofuli,

Good render time,


I think it is the only way to render a scene like this one:
_adjust exposure (not too bright, brightness means longer render times )
_work with smooth filtering ( time consuming but very efficient)
_adjust the glossiness ( too low is very noisy) and adjust the subdivison based on the glossiness like you did.
_"use light cache with glossy ray" ticked( in case of big glossiness)

Overall i think it is better to work locally instead of trying to manage everything all together ( universal setting for example) . Joconell written some very good advice about it in some threads like this one :
http://www.chaosgroup.com/forums/vbu...ng-DMC-Sampler
It is very helpfull to understand how to adapt each scene because there is no "magic recipe" unfortunately!

regard's

[^Lele^]

I Lele,

the goal is to reduce the rendering time with a good noise with BF+LC with sub-pixel OFF

No, I Lele! ^^

I think what i meant to say is that the scene is set up wrong, so the benchmark isn't a benchmark at all, if not of how VRay is doing at tracing the wrong rays to a clean state.
It has no bearing whatsoever with a correct approach to a scene setup, and it misses the point off the bat by not being linear in respect to anything.
That is as capital a sin as I can imagine, these days, and voids any other step that can be taken further down the line (i'm not debating personal preferences here, understand. there are a number of very good reasons for a linear workflow, none of which are down to opinion, but rather to numbers).
The shading, modelling, and settings issues just add to it.
Looks more like a redoing than a fixing, to me.

If you provided a clean scene (without plugins, like quadchamfer, but with static meshes and proxies of correct topology), with a proper linear lighting/colormapping setup, and wanted to explore settings without changing the look of the result, that could be done nicely in such a post.
If you provided the aforementioned clean scene, but with your current rehinard mapping, and asked how the scene ought to be set up to be "proper", regardless of the current visual results, that could be done.
You would likely also get a few ideas on how to speed up the process of delivering such images to the client, through compositing.
Both at the same time, not quite: it will keep rendering this slow, if not slower, and won't clean up, the more tracing finding more direct reflections of bright light fixtures, needing more samples to clean those, which in turn will generate more, and so on, and so forth.
The changes to stop the noise from spreading in a reasonable time are so radical in setup (if a line or two of scripting, again), the colormapping approach so different, that the picture will not look the same.
You'd be able to match most of it in post, however not all of it, as your colormapping, the way it's done, is going to skew the GI (additively, bounce by bounce), which you won't be able to recover in post (eyeball, maybe. exactly, no.)

Of course, feel free to do whatever it is you wish, just don't label the renderer as slow, or lay it on VRay's doorstep, as i honestly don't think its for the renderer to fix, in this case: i'd say given the number of rays it's asked to trace, adaptivity here is gifting you with a render at all.
With other engines, you'd still be waiting for the first bucket to move, a day later (and no, it ain't hyperbole either.).
If it wasn't because of cutoffs defaulting @ 0.01 in the shaders and lights, and the adaptivity of the sampler, a 5th bounce glossy with 128 subdivs would trace a 1180591620717411300000 rays, or (128^2)^5, per pixel, or a Sextillion rays per pixel, in U.S. money.
Go figure what happens in your black walls corners, and add to that the high intensity direct reflections, and you get the idea.

I suppose the sum of my contribution here is: go linear.
I'll leave you to it!

[glorybound]

So, from 4 hrs to 10 minutes! Holy Moley, batman. I am printing this post out, and pasting it to my forehead!

[cecofuli]

Hi cecofuli,

Good render time,


I think it is the only way to render a scene like this one:
_adjust exposure (not too bright, brightness means longer render times )
_work with smooth filtering ( time consuming but very efficient)
_adjust the glossiness ( too low is very noisy) and adjust the subdivison based on the glossiness like you did.
_"use light cache with glossy ray" ticked( in case of big glossiness)

Overall i think it is better to work locally instead of trying to manage everything all together ( universal setting for example) . Joconell written some very good advice about it in some threads like this one :
http://www.chaosgroup.com/forums/vbu...ng-DMC-Sampler
It is very helpfull to understand how to adapt each scene because there is no "magic recipe" unfortunately!

regard's

This is an incredible thread. I missed it! I tried it and, at the end, I have similar value as my last test ^__^'
AA 4 16 is the best solution for this interior. I have to use low subdivs for trees and hight subdivs for wall and ceiling.
I don't know what I can do. If you have suggestion, I'm here. Otherwise, I upload my optimized scene.

[Kr0no]

No, I Lele! ^^

I think what i meant to say is that the scene is set up wrong, so the benchmark isn't a benchmark at all, if not of how VRay is doing at tracing the wrong rays to a clean state.
It has no bearing whatsoever with a correct approach to a scene setup, and it misses the point off the bat by not being linear in respect to anything.
That is as capital a sin as I can imagine, these days, and voids any other step that can be taken further down the line (i'm not debating personal preferences here, understand. there are a number of very good reasons for a linear workflow, none of which are down to opinion, but rather to numbers).
The shading, modelling, and settings issues just add to it.
Looks more like a redoing than a fixing, to me.

If you provided a clean scene (without plugins, like quadchamfer, but with static meshes and proxies of correct topology), with a proper linear lighting/colormapping setup, and wanted to explore settings without changing the look of the result, that could be done nicely in such a post.
If you provided the aforementioned clean scene, but with your current rehinard mapping, and asked how the scene ought to be set up to be "proper", regardless of the current visual results, that could be done.
You would likely also get a few ideas on how to speed up the process of delivering such images to the client, through compositing.
Both at the same time, not quite: it will keep rendering this slow, if not slower, and won't clean up, the more tracing finding more direct reflections of bright light fixtures, needing more samples to clean those, which in turn will generate more, and so on, and so forth.
The changes to stop the noise from spreading in a reasonable time are so radical in setup (if a line or two of scripting, again), the colormapping approach so different, that the picture will not look the same.
You'd be able to match most of it in post, however not all of it, as your colormapping, the way it's done, is going to skew the GI (additively, bounce by bounce), which you won't be able to recover in post (eyeball, maybe. exactly, no.)

Of course, feel free to do whatever it is you wish, just don't label the renderer as slow, or lay it on VRay's doorstep, as i honestly don't think its for the renderer to fix, in this case: i'd say given the number of rays it's asked to trace, adaptivity here is gifting you with a render at all.
With other engines, you'd still be waiting for the first bucket to move, a day later (and no, it ain't hyperbole either.).
If it wasn't because of cutoffs defaulting @ 0.01 in the shaders and lights, and the adaptivity of the sampler, a 5th bounce glossy with 128 subdivs would trace a 1180591620717411300000 rays, or (128^2)^5, per pixel, or a Sextillion rays per pixel, in U.S. money.
Go figure what happens in your black walls corners, and add to that the high intensity direct reflections, and you get the idea.

I suppose the sum of my contribution here is: go linear.
I'll leave you to it!

Amen to that!

[cecofuli]

I upload a new version, I hope, with better setting (1h 50min). Same link as the old one.
Only to change the file name from 01_cecofuli_test.rar to 02_cecofuli_test.rar
If you have suggestion, we are here ^__^'

[vlado]

There is a bug in the scene that took me the longest time to find and I thought I was going mad... the window panes are missing one plane - there are only 3 planes instead of four. Would be nice if you can correct the scene, especially if you plan to use it for benchmarking...

Best regards,
Vlado

[cecofuli]

XD O my god! And, this was an evermotion model. I didn't know this problem!

[^Lele^]

Vlado, my "short" post above mentioned it :P

"i believe that's just removing the floating poly sheet on the outside of the windows."

[^Lele^]

XD O my god! And, this was an evermotion model. I didn't know this problem!

Oh, that's rich.

[cecofuli]

I don't remember, but I think in my last scene I removed the double panel.
When we are in a short delivery, deadline, to use a pre-model objects is a common way.
Strange that Evermotion did this error...

[vlado]

I only noticed it because it was causing differences in the rendering in the 3.0 builds in some cases...

Best regards,
Vlado

[^Lele^]

I don't remember, but I think in my last scene I removed the double panel.
When we are in a short delivery, deadline, to use a pre-model objects is a common way.
Strange that Evermotion did this error...

I rest the case I made in my previous posts.