A return to equal priority on render speed?

[zero-13]

I thought I was going crazy but after much testing and tons of confirmation online I have sadly accepted that V-Ray 5 is undeniably slower than V-Ray Next was with equivalent settings.

I don’t understand why this has to be, for example V-Ray Next GPU settings were almost identical to current settings save for the big differences like GPU LC and trimming a few older settings etc.

The reasoning in all the threads I’ve read is that a higher priority has been placed on consistent quality at the cost of some speed, I imagine this means overall samples are higher than what similar settings in V-Ray Next and prior display compared to equivalent V-Ray 5 settings.

V-Ray as far back as I can remember has always improved performance in all aspects especially render speed until version 5 and I have not read a truly compelling reason as to the loss in performance over difficult to perceive fidelity comparatively speaking, certainly ease of use is not a very good reason to slow a render engine down… as we’ve had the option between Simple, Advanced and Expert settings for a long while now.

Please understand this is more confused sadness than complaint, I think V-Ray 5 is wonderful, the additions, improvements and ease of use have all moved forward, but why can’t the old loyal Chaos Group standards of always improving speed as well as quality be rekindled.

The less render power you have the more vital and welcome speed increases become, for studios with render farms speed decreases can be negligible but for Individual artists with one or two GPUs or less than 100s of cores it’s kinda brutal.

I’d much rather update to a newer build than have to revert to an older build to regain a speed advantage.

I just hope that render speed which has been one of V-Ray’s legacy hallmarks does not continue to be sacrificed as time moves forward.

V-Ray = Speed + Quality

[^Lele^]

Are you referring to the GPU engine?
Also, it'd be very beneficial for us if you could provide us with scenes, so that we could figure out where the speed is lost.

[zero-13]

Hi Lele, I have noticed with GPU mainly, only ever switch to CPU if I occasionally have an issue where GPU does something weird as my dual 12-core Xeons are usually slower than my single 2080 Ti.

As far as GPU settings the only parameters ever tweaked are noise limit and samples, and both RTX and CUDA seem similar with CUDA occasionaly performing a bit better than RTX which is also new and odd as RTX has always been significantly faster prior to V-Ray 5.

Apologies I don’t have the links right now as I reply, but there are several threads here and a few external like reddit discussing the slower overall performance of V-Ray 5 versus earlier versions of V-Ray Next etc.

As mentioned, every prior update I’ve ever installed prior to 5 has provided an increase in render speeds most noticeable once V-Ray RT became V-Ray GPU etc. so after having installed V-Ray 5 and using it a bit on new scenes it felt slower but it didn’t occur to me that it actually might be as I’ve had that expectation of either an increase of performance, a somewhat lateral move but never a decrease.

You know how most of us get a sense over time about how long an animation sequence might take just from experience and based on scene materials animation etc.

So my last animation sequence taking much longer than expected finally sent me on the hunt and so I found several people experiencing slower performance and from those replies it seems somewhat known.

So I’m just generally bringing it up as a noticeable performance decrease for V-Ray not related to a hot fix etc. is a first for me and so just being a squeaky wheel.

My plan today is to test and reacquire the minimum number of sample & noise threshold to compensate for whatever new increased sampling has been implemented if possible, it would be great to know what has been cranked up under the hood so that I might get back to a baseline speed / fidelity.

Seems a better choice than down grading just to utilize old tried and true settings.

Apologies on the Wall-o-text, just a few nerd tears being shared.

[^Lele^]

I hear you, no worries!

GPU is an odd beast: it depends on a number of factors outside of our control.
If you reverted to a previous version, with a current driver, you'd perhaps still experience slower rendertimes.
Sticking to older driver versions isn't really possible, as new tech gets developed all the time, and that requires updating, chasing those changes and more, nearly as often as a driver gets updated.

We can't, and rather not by choice, control narratives from users, much less so on reddits or other such media.
It's all too easy to find confirmations to one's biases (and here i'm talking purely about myself.) when doing experiments, tests, and benchmarks.
I haven't been under production pressure for a while now, but under duress, one is bound to try and find the quickest solution to a problem: we accept the inevitability of blame laying at our feet, unpleasant, and motivated or not, as that may be.

I can say two things only with certainty:

a) We have not measured the same slowdowns as others do, even when given scenes that apparently showed it: they simply never reproduced on our side, on multiple workstation types, with multiple test conducted by different people (me, QA, support, the devs...).
We strive to enhance performance across the board, and no new feature is ever thrown into the mix without rigorous assessment of its impact on speed and quality. When we get that balance wrong, as we're bound to from time to time, we really would like to know, so to correct the situation.
Please provide us with scenes showing undue behaviour, as what is obvious to the user often isn't when looked at from this side.

b) We're only too happy to help our users to get the best out of V-Ray: pack and send us a scene to get checked, if you feel it's slow in rendering, and we'll try our best to give it back rendering quicker.
If not, we may still provide for insight on what is causing slowdowns on the setup side, and perhaps even leave the users with a few tips on how to control those quirks by themselves.
So feel free to share the burden as early as you can, and we'll be here trying to make things better.

As for what changed, well, a lot has in a number of key departments.
We make no promise that *old scenes* will render faster, but rather that a *given feature* will be quicker, as the versions go up.
As the scenes' setups *are changed* to accomodate the new approaches, those too will become quicker.
When one sees a video from us comparing two versions side-by-side, one can be sure that there had been changes under the hood so to leverage the new methods and features (f.e. removing blends to use the vrayMtl extra specular layers, removing portals to use the adaptive dome, etc.), unless of course that change isn't required (f.e. an update on embree, pre-render, or on the denoiser, post-render.).
Perhaps we should be clearer on this part, in the future.

Wall text for wall text, i hope this is received in the same kind as it was given.

Also, happy festivities!

[zero-13]

Thank you for the thoughtful detailed reply! I definitely understand and am reminded how the GPU drivers can greatly affect and even destroy performance and compatibility as I’ve experienced in the past.

I think that’s testimony that things have been fast and stable for so long with GPU, that perhaps I too easily let those old memories fade.

I also realize with the bump in quality I should be able to tweak samples and noise to get speed back to where it was.

I appreciate everyone’s hard work, when I think back on what I used to think were blazing fast render times on CPU compared to current GPU speeds I start to feel spoiled lol.

Again, thank you for the reminder that the constant changes to GPU drivers can add much variance and difficulty outside of the team’s control with the constant changes vs CPU which is inherently more consistent.