V-Ray Scene Optimisation Script

[Paul Oblomov]

If i remember correctly, I turned off generate GI and all the stuff in emitting plane

[^Lele^]

Ah, through obj props.
Never checked those, two ways of skin the same cat.

[^Lele^]

I had another go at it, given the noisy spec/reflection and the thin lines not being quite well resolved without image filtering.
Further, i tried to better match the original look of the wood planks and woodden bookshelf on the left, by the sofa (minor minor tweaks to reflection Ior and maps)
For whatever reason, the nightly i am using isn't accepting the frame stamp, so please forgive me for omitting it, and stating the render time here.
This took 1h 8min 49.6sec
I used the VFX-Cmplx preset, and modified it in two places:
AA was adaptive 8/8 (same as fixed 8 ) down from 16/16, base subdivs remained set to 2, and the GSM was 24, to achieve 2306 SPP across all effects.

the scene is available here (no maps, as they didn't change from what Paul provided in the first place)
https://drive.google.com/file/d/0B1P...it?usp=sharing
The thick exr is here:
https://drive.google.com/file/d/0B1P...it?usp=sharing
Sample rate is omitted from the images below for obvious reasons.

[^Lele^]

To share with you the analysis of what AlexP sent me in private, in as honest and unbiased a way I can, i'll post here what i wrote to him.

Umh, yes. with DoF, you will need more camera rays, definitely, than 4 (or 6, for that matter).
The approach is VERY similar to the MSR one, in this case, but i'd start from the minimum amount of camera rays needed to give you a clean Dof, and fix those (rather than leaving the AA make its choices and potentially run off), and then raise the SPP to match.
Perhaps the VFX-Cmplx, which has 16 cam rays, may give you better results and clean the DoF noise, but hey, it's of course dependent from how much the DoF will spread them.
In general, Camera rays (as per the three tech posts of yesterday) are around 10 times slower than specialised ones, so they'd have to be used as sparingly as one possibly could.
This said, there are shots for which no cure exists but Cam rays, and a long wait, and yours look to be precisely in that league: the gains WILL be there with my methods, but perhaps so marginal that rendering problematic regions to find the ideal balance wastes you more time than having it rendered with a MSR/Universal and wait it out (MSR seems the best approach by far at 1h20mins).

So, as stated before, Vlado DOES have it right, for the vast majority of cases, with the approach of raising the "min shading rate" (which to all effects is similar to my script/method, if it doesn't quite allow for fine numerical control over the samples that will be cast).
That too, after all, allows the feeding of a cleaner shading component to the AA, reducing the need to cast more expensive camera rays.

[kosso_olli]

Lele, is this meant to be used for high res print work? I tried it out, but no matter what preset I choose, the buckets are stuck in the top left corner of the image since half an hour... The only setting that works is the "rough" one, which ends up looking just like that.

[^Lele^]

Lele, is this meant to be used for high res print work? I tried it out, but no matter what preset I choose, the buckets are stuck in the top left corner of the image since half an hour... The only setting that works is the "rough" one, which ends up looking just like that.

Yes, resolution does not matter.
however, if you started from a non BF/LC GI setup, the rendertime increase will be quite severe (ie. an IRMap is much quicker).
Was that the case?
If not, more details will help me figure out what's happening.
feel free to send me stuff in private, should you wish so.

[Paul Oblomov]

Freaking hour. OMG
Will try to render fullHD tonight

[^Lele^]

Eh, as i said, it's THREE times slower than the average 24 core intel i7.
The test which took 1hr1minute here took 22 minutes on one of those (remeber in the beta group?).
I need a new one, but i loathe the silly intel pricing, so i'll wait for a bit longer to switch.
In the while, yeah, PLEASE bring the times to those normal in 2014, lol!
Thanks for that, Paul.

[kosso_olli]

Yes, resolution does not matter.
however, if you started from a non BF/LC GI setup, the rendertime increase will be quite severe (ie. an IRMap is much quicker).
Was that the case?
If not, more details will help me figure out what's happening.
feel free to send me stuff in private, should you wish so.

Can certainly do that, although with a stripped down scene. But you should get an idea of what is going on.
I came from a kind of universal bf/lc approach suggested by Vlado, using the MSR parameter to clear things up. The scene does use DOF and displacement, maybe that is the problem. Where can I send the scene?

[vlado]

The test which took 1hr1minute here took 22 minutes on one of those (remeber in the beta group?). I need a new one, but i loathe the silly intel pricing, so i'll wait for a bit longer to switch. In the while, yeah, PLEASE bring the times to those normal in 2014, lol!

An hour still seems way too slow for what V-Ray can do with this scene, even for your hardware. Just so you know what you are up against, Corona renders this in a few minutes on a regular i7. V-Ray generally performs slightly slower from my own tests. But I don't know why it's taking so long with your setup.

Best regards,
Vlado

[^Lele^]

oh hey there!
I've posted the scene on purpose, of course, so that others could test it.
We're still missing one key ingredient to fully unbiased benchmarking: that (in)famous rnder element i asked for a while back which would express the pixel variance.
This renders surprisingly quick on my rig as well, if i do not look at noise levels at 800% zoom.
I have yet to see a comparison, though, with either engine, and elements this clean.
If we could take the subjective perception of noise (as my slow mxs tool did, or a faster nuke gizmo did), and express that as a straight number, then perhaps i'd figure out why.

[^Lele^]

Can certainly do that, although with a stripped down scene. But you should get an idea of what is going on.
I came from a kind of universal bf/lc approach suggested by Vlado, using the MSR parameter to clear things up. The scene does use DOF and displacement, maybe that is the problem. Where can I send the scene?

I think google drive or equivalent is ideal.
you can send me the link in private, and delete the scene as i downloaded it.
Gmail may have attachment limits...

[kosso_olli]

Thanks. PM with the link is sent.

[joconnell]

Here are the images i tested the archviz preset with.
The rendertime on my puny AMD FX8350 @ 4.4 Ghz was 1h30mins.

The scene and maps will follow in further posts for anybody interested to give it a try with various setup approaches.

[ATTACH=CONFIG]20448[/ATTACH]

Your plant sucks.

[^Lele^]

Thank YOU!
Downloaded and replied. you can safely delete it from where you uploaded it.