Vray 5 to 3.7 compatability

[busseynova]

I know I could just try this, but I'm a bit scared to go uninstalling and installing various version whilst I'm in the middle of a project. I upgraded to the trial version of Vray 5, driven by the need to render on a farm that runs that verison. I like it and will try and convince my boss that it's a worthy expenditure but.... If this is not possible and I have to downgrade, will I be able to open the Vray 5 scene in Vray 3.7?

I understand there will be a caveat that some things may render differently / incorrectly, but as long as I can open the file in order to adjust these back to something that works in 3.7, I think that will be ok.

[^Lele^]

Yes, you should have no issue whatsoever (i have done this a few times over, of late.), with the caveats you well have in mind.

[busseynova]

Thanks for confirming. I'm actually going to try installing 3.7 on max 2021 so I can do some side-by-side speed tests

[^Lele^]

Ensure you bring the proper techs to bear before doing so (f.e. adaptive dome.).
Also, verify the rendered image noise (*not* N.T.!) matches: the newer V-Ray will generally do more work to give you a cleaner image, where 3.7 happily left some noise, sometimes at the expense of some rendertime.
It generally more than turns out as favourable anyway, but mileage may vary.

[busseynova]

Ensure you bring the proper techs to bear before doing so (f.e. adaptive dome.).
Also, verify the rendered image noise (*not* N.T.!) matches: the newer V-Ray will generally do more work to give you a cleaner image, where 3.7 happily left some noise, sometimes at the expense of some rendertime.
It generally more than turns out as favourable anyway, but mileage may vary.

How would you verify that the noise is the same? Just by visually inspecting the noiselevels RE?

[^Lele^]

I built myself a gizmo in Nuke to do that.
A 9 pixel kernel checks the central pixel against the average of the surrounding ones.
You could, however, use any other tool (i think there are image viewers that can do that. EDIT: perhaps neat image has been used by others in the forums to do this.), provided you measured the images in the same way.
In other words, the math isn't so important, provided it's the same across tested images.

[busseynova]

Hmm, I actually get 39 seconds longer in Vray5 on the attached scene (bucket) but maybe I'm setting it up wrong. The noise levels RE seems totally clean, I just used the default settings and only changed res and image sampler.

Is there such a thing as a vray benchmark scene?

[^Lele^]

The noise level will always look "totally clean", given it may contain values below the display abilities of a standard 8bit, sRGB monitor (at a default of 0.01, that's a value of 2.55 or lower, or 2, in 8 bit integer money.).
You'll want to at least get an average of the float values of the pixels in that RE.

In any event, try a scene which is production grade, once you know you're making both versions do the same work.

I can't download your attachment, for whatever reason.

Benchmarking is a dark art, too often misinterpreted.
Meaning it can be used, and we do use it, to gauge *specific* aspects of a render engine (meaning us, and others too.) under extremely controlled conditions, but never to assess its overall goodness (or, speed in converging to a given noise threshold in a production scene.).
As such, you won't find us comparing this to that in a simplistic fashion, nor we encourage doing so, unless one was exceedingly certain of the quality of the task brought to bear (in which case, one would know why it's impossible to compare, eh.).

Likely, v5 cleaned the image, perhaps below visible threshold, in a number of places v3.x skipped.
We've seen this time and again, due to improved algorithms, and it wouldn't be surprising to see it happen again.
Alas, do share your findings.

[busseynova]

I'm not sure what you mean by production grade, but I just tested a production scene, granted it uses vray sun/sky in a max daylight system not a dome light.

The Vray 3 took 75% of the rime to render and the averaged floating point noise level is slightly lower too. Some of this may be (probably is) due to bad setup in my scene, but I'm struggling to see how that should result in a drop in comparative performance from one to the other?

I'd be interested to see a scene where 5 is proved to be faster/better. I like the lightmix tool (I was hacking this previously with a lightselect script), but I'm not seeing that it'll be worth the upgrade at the moment.

[^Lele^]

https://forums.chaosgroup.com/forum/...getting-slower

[busseynova]

Ok, fair enough, so Vray5 renders better if not faster under some circumstances. That's a harder sell to my bosses, the light Cache improvements do sound good though.

[^Lele^]

When i said "use a production scene", and "ensure you activate the new tech" i meant it precisely to have you have a better time with the transition, and make the sell easier.
You will want the adaptive dome active (so, no background, for now. the map needs to sit in a dome.), and in a complex animation with trees moving in the wind, casting shadows inside an interior, for example, the new LC would fly through, per frame, without you giving it a second thought, and without producing artifacts, and avoiding the painful, and error-prone, save and reload of LC sequences.

On the shaders' side, the speed improvements are not marginal, when compared to identical shaders built with the blend material, and also rendertimes will at all times stay manageable because using the VrayMtl only will ensure energy preservation.

Think of the time saving of the light mixer, and when needed, LPEs.

Think of the new VFB, now and as it'll improve across v5 and later, as a handy compositing tool.

Clocking ray for ray is a poor metric (although, there too we improve), as the conditions in a complex scene will make the ray *count* very variable (to ensure you get a uniform amount of noise across your render).

[busseynova]

Well Vray is pretty fast anyway granted, compared to Corona for example. Having used the lightmix briefly it definitely has the potential to speed up workflow a lot too. Reading a colour management thread I want to learn more about that too as it has the potential to let me do even less in post.

compared to identical shaders built with the blend material

Can you clear up what you mean by this? I guess those new coat in Vraymtl options are a better alternative to the blend material?

have you set out all these new best practices anywhere in one place so I can stop asking questions you've probably answered 1000 times?