V-Ray Scene Optimisation Script

[^Lele^]

Obv all scenes are different, but I dont understand why this way of sampling is faster than say, just upping subdivs on lights & materials or upping MSR.

Well, you surely could do that, however, without turning "divide shading subdivs" off, the max AA value will divide the subdivs you set by itself (ie. 8 subdivs on a light, and 8 max AA will result in 1 subdiv times the min AA subdivs.)
If the min AA subdivs are 1, then you get the first "pass" to have only one subdiv for the light, hence it becomes noisy, hence the AA is forced to step up (which as per my posted test above is slow-ish).
Same goes for the MSR: say you want to counter the division of shading subdivs, then your MSR ought to be 8 (GSM being 1.0).
However, that leaves you with a bit less control over what happens at the MIN AA subdiv level, so you have to rely more on the noise threshold for the AA to do the right thing.

Additionally I dont understand why 4/6 AA isnt much slower. Also, I dont understand why the egdes are clean with clr thresh @ 0.1

Much slower than what else? :P
The Alpha edges are clean because the MIN AA subdivs is set to 4, so i force VRay to trace 16 camera rays for the heck of it, however, given the "underlying" scene is clean (through decoupling the AA and the global DMC thresholds, with the latter MUCH lower than the former), those camera rays do not "mutate" into the specialsed ones, but work purely for the Geo, and little else.
That's why they are quicker.

I suppose the same results could ideally be achieved by the decoupling of the thresholds, a fixed (or near fixed) AA sampling, and the use of the MSR spinner, but hey, when i started this script, very early on in the beta of 3, there was no MSR :P
Also, it's a method that i brought over from vray 2.x, where i had to resort to the adaptive subdivision AA and the GSM to achieve similarly quick results (although now it's a wee bit quicker, this way).
Feel free to try out the other approach, i'd be curious to see a benchmark!

[AlexP]

Everytime I think Ive got a good understanding of rendering with Vray, something else comes up.

Will do some tests with different setups a bit more scientifically, but on first look using the *arch viz* preset has given me exactly what I want (other than turning on MRI)

ps. Would it be possible to add an option to leave DR on? I keep forgetting and then send renders to BB with DR turned off. Or can I remove that part from the script? (I haven't looked)

[^Lele^]

Everytime I think Ive got a good understanding of rendering with Vray, something else comes up.

Eh, tell me about it.
I meant to write a sampling guide for the best part of six months, now, and every time i start, i run into more questions.
Vlado very kindly answers those, and as a result, i am thrown into the arms of a dozen more...
Then again, i wouldn't want it any way else.
It's the salt which gives true flavour to what i do for a living, i guess

Will do some tests with different setups a bit more scientifically, but on first look using the *arch viz* preset has given me exactly what I want (other than turning on MRI)

Was it only cleaner, or also faster than your previous setup?
Or you just had no previous setup to begin with?

Would it be possible to add an option to leave DR on? I keep forgetting and then send renders to BB with DR turned off. Or can I remove that part from the script? (I haven't looked)

I guess i may just leave the DR part alone.
Never having used it in production, the function in the struct file which reads and writes the render settings didn't have a specialised part for it.
Will do so as i get the new version rolled out.

Thank you ever so much for the feedback: it's gold to me!

[jacksc02]

Lele,
A DR on/off switch would be handy.
Is 2 BF GI subdivs "normal" for VFX scenes?

[^Lele^]

sorry for having disappeared, all of a sudden.
will be fully operative this evening, or tomorrow at the latest...

yeah Chris, in vfx, noise is THE next thing, eheh
jokes aside, it's 2 subdivs, times the GSM, times the adaptive fraction, times the AA subdivs.
i think it's 768 SPP.
check the lower info tab for the exact number (I'm trying to remember, and you know how that works -not- for me...)

EDIT: i was typing from an asus tablet. Worst typing experience ever.

[Neilg]

i'm not getting great results with this.

I lowered my res to 800x800 and did some tests.
Top one is my settings - I always start my renders with default, change AA to 2,8, start with all subdivisions at 8 and increase as required.
second one is your clean preset, one after that is 2stops.

[MoonDoggie]

Cubicle? Did you get a new name?

[Neilg]

yeah, I was a child 10 years ago when I signed up, time to use my real name

[^Lele^]

clean and 2-stops are the oldest ones, as i stated in the first post.

The various pre-made presets were tests, really. I'd recommend using the "StdGeo" and "FineGeo" ones for best results, and in case save yourself a couple more presets with the global subdivs multiplier halved, should you have scenes which tend to clean up quicker (ie. aim for 512 SPP on all effects).

Besides this, it'd be interesting to see where the noise is coming from ( i have a feeling it's from GI).
What types of GI you were using for your original render (if you can get 17 minutes of clean BF GI for an interior, while a 24 subdivs takes three times longer, and a 30 subdivs takes six times longer, you have some tricks up your sleeve)?
Also, as stated above, the current version of the script is turning off max ray intensity, normally on by default, which WILL produce uncleanable noise if you're lighting with extremely bright sources.
Further, your settings seem to have an area filter, which masks low-intensity noise a bit more (but makes a hellish job for compers in a number of scenarios), while filtering of the render is something my settings turn off.

Any info you can provide me with in terms of scene setup would help figuring out why some are getting 60% speeds increases, and in your case there's a 600% slowdown.

[Neilg]

It was all IR/LC - there's no noise in the GI. and this is the first render i've done in ages where i forgot to turn off the filter. whoops!
There are 2 light sources - a sun, which just about burns out (intensity of 0.075, ends up with values slightly over 1 in the render) and a vraysky at 0.3.

Your settings drop the material subdivisions a lot - my glossy surfaces range between 32-64. I always leave the noise threshold at 0.01 and yours goes crazy low, which accounts for the rendertimes but somehow doesn't clean up the image. I had .8 on my reinhard burn (calculation only) which made a huge difference rendering the transparent curtains hit from behind by the sun.
I cant see how the IR/LC changed at all between them, but those passes were much slower too with no visible change in quality.
There's about the same level of noise in the reflection pass between the first 2, the third is cleaner (although needlessly so) and it's mostly the AA and the edges. If i run your preset and turn it up to 2,8 it barely gets started though.
I think the increase in render time is coming from the super low noise threshold, but there aren't enough AA or material samples anywhere for it to actually clean it up.

I'd wager the reason everyone else is getting speed increases is because they're coming from universal settings.

[^Lele^]

Well, we're comparing oranges to apples then
Not everyone comes for a universal settings scenario, but to have a comparison, you want to use BF/LC, and whatever render settings you please.
Two posts you may have missed:
http://forums.chaosgroup.com/showthr...384#post627384
on why the script WILL NOT work with IRMAP.
http://forums.chaosgroup.com/showthr...505#post627505
On how the math behind the interface works, why the subdivs settings LOOK deceptively low, and what actual sampling VRay's requested to do.
The info panel on the script, too, is meant to be read to give the user a queue as to what to expect from the renderer.
A more detailed explanation of the concepts behind it (which you may find useful, or not. it's not a one-size-fits-all situation...) can be found here:
http://forums.chaosgroup.com/showthr...414#post627414
and here:
http://forums.chaosgroup.com/showthr...498#post627498

Further, on why ONLY a LINEAR colormapping option is meant to be used with the script is here:
http://forums.chaosgroup.com/showthr...394#post627394

If an IRMAP floats your boat, then the "divide shading subdivs" option off, and what the rest of the script does to the scene leveraging that, will likely sink it, i'm afraid. ^^

[Neilg]

What if the IR/LC is saved using different settings and loaded? we're still talking 10 minutes to 30 (or 1hr 15) with the same levels of noise.

[AlexP]

FWIW I had great results on a couple of exteriors - using BF/LC & dome & hdri

But have been testing on an interior and had the opposite, closer to what you had Neilg. (but using dome & hdri)
Mainly because the BF can't compete with the IR map & *normal* settings (basically default Vray with a couple of tweaks)

I was also unable to find a way to successfully clean the DOF on the interior using the script.

Im playing with ignoring the GI settings and using only the AA & DMC settings from the script but I think everything is finely balanced to work together.

I was planning to do a couple of renders at hi res over the weekend as tests with different setups, I'll send you the results Lele if helpful?

But I wonder if this approach just isn't really suited to the typically challenging GI situation in interiors, just as a guess, Im not an expert

[^Lele^]

Well, I have not used interpolated methods for primary GI in years now.
So, sure, if the desire is to use one, because the primarily visible surfaces are big, flat and even, those methods will always come up on top for render time than a multi-sample-per-pixel, BF GI : it's not a new finding, as you well pointed out, Alex .
But interpolated methods have a set of issues and setup caveats of their own, and they cannot be reliably expected to work (consistently, either speed or quality-wise) across sequences which are geometrically diverse, and throwing away a long-rendering sequence because of it it's unacceptable, in my book (throw away a night on a big farm to render a 300 frames, 2k shot in ANY company to get botched results, and you will positively be hearing yelling. That's called "Rendering to the thrash-can", and no LnR artist or TD would want to hear it associated to their names, as it doesn't really do one's credit any good.).
For example, it's sure very easy to set up an IRMap for a still, or a slow camera pull of the same room.
But if you geometric detail frequency changes across the camera move, and for some reason you find a wall like the one to the screen right of the image i posted (the scene is from John O'Connell, and I suppose I could ask him to share it as a tester, once properly cleaned and once made sure of no copyright/licensing infringement), then interpolated methods would, and will, fall apart.
Less noticeably in a still, but much more so in a moving camera, or god forbid, moving or deforming geometry.

We are talking of EXACT approaches to rendering, which will ALWAYS return a viable sequence, no matter the contents.
ArchViz is not the only place where interiors are rendered, only in VFX you also have to marry that with people, or destruction, or whatyouhavenot.
IRMap would die before the word go.
What i stated, unfortunately, remains valid, for this approach:

If an IRMAP floats your boat, then the "divide shading subdivs" option off, and what the rest of the script does to the scene leveraging that, will likely sink it, i'm afraid. ^^

Moving back on topic:

I was also unable to find a way to successfully clean the DOF on the interior using the script.

This could well be the case, although the VFX-Cmplx preset could help.
Of course, the amount of DoF and/or MoBlur is unknown to begin with, but the method shows the way, and the script enables you to execute it: since in v3 the two aforementioned effects are wholly governed by the camera rays set in the AA page, you should grab the worst frame for either DoF or MoBlur, set the AA rays high enough so that the alpha channel is clean for the objects outline, and use the other spinners (GSM, chiefly) to MAINTAIN the samples per pixel at the level which you find acceptable for your RGB render and individual shading effects, without fear of breaking down the all-important (for compers all the more so) alpha edges.
You will then be able to right-mouse click one of the custom preset buttons, and save out the one that works with a name of your choice.

Im playing with ignoring the GI settings and using only the AA & DMC settings from the script but I think everything is finely balanced to work together.

Eh, yes, unfortunately.
Doing what i do to the GSM, noise thresholds and so on is meant to work at unison, or fall apart.
As i stated before, it's meant for EXTREME optimisation, and to achieve these kind of speed bumps, especially without knowledge of the content of a scene, means the logic structure is fairly intricate, and not very prone to fiddling with (not the script, but rather the underlying logic. Marry for the life of a shot/sequence, or ignore. There's hardly the chance for one-night stands )

I was planning to do a couple of renders at hi res over the weekend as tests with different setups, I'll send you the results Lele if helpful?

PLEASE!
Either here, or in private, as you prefer.
I'm more than happy to discuss and debate them in public: i would have kept the script to the closed beta group if i wasn't prepared to have it ripped apart

But I wonder if this approach just isn't really suited to the typically challenging GI situation in interiors, just as a guess, Im not an expert

Quite the contrary, as you may have evinced from the first part in this post related to interpolated methods.
Get geo detail which is sub-pixel, render me an IRMap render of it, and show me your GI render element, and i will show you where it breaks.
You'll end up pushing the IRMap so hard that it will become slower than a BF approach, and still produce issues.
I have tried, ages ago, so i know first-hand.

[AlexP]

re: the DOF I guess what i meant to say was, I wasn't sure the right way to manipulate the settings to clean the DOF. Will try as you've said.

All interesting stuff.