Max subdivs not matching the amount of passes

[LudvikKoutny]

I did not measure the noise in any way. I never encountered that issue myself, as in Corona, I always rendered for a fixed amount of passes, but people who did encounter it, did not find about it doing some analytic noise measurement. They simply saw that noise amount flickers from frame to frame, and it's visible and unpleasant to their eyes. If it was not obvious to a naked eye, then there would be no reason to fix it in the first place.

I saw your guide actually, and I apply it pretty much day to day. I am not sure you are fully in context here. It's about single scene that renders very slowly even with very high noise threshold, like 0.2. Vast majority of my other scenes render perfectly and quickly in V-Ray. And this particular thread was just a question if there is any reason that Max. subdivs do not match the amount of passes, which was already answered.

[anthonyh]

ooh, you mentioned the `C` word ! heehee
9)

[^Lele^]

I did not measure the noise in any way. I never encountered that issue myself, as in Corona, I always rendered for a fixed amount of passes, but people who did encounter it, did not find about it doing some analytic noise measurement. They simply saw that noise amount flickers from frame to frame, and it's visible and unpleasant to their eyes. If it was not obvious to a naked eye, then there would be no reason to fix it in the first place.

No numbers, no party.
Without a numeric value for noise threshold, the number of variable in play into the perceptual domain are nearly infinte.
Time is the first one: a monitor loses calibration while time passes, its LEDs lose emission capacity, and so on and so forth. I am assuming you work under a calibrated sRGB monitor, ofc, otherwise throw in the display device as another variable.
Weather is the second: eye perception changes with lighting conditions.
Emotions are the third: stress and pressure will bias judgement (no, really.).
The list goes on and on, in fact, and that is why "perceptual" is by definition "biased".

I saw your guide actually, and I apply it pretty much day to day. I am not sure you are fully in context here. It's about single scene that renders very slowly even with very high noise threshold, like 0.2. Vast majority of my other scenes render perfectly and quickly in V-Ray. And this particular thread was just a question if there is any reason that Max. subdivs do not match the amount of passes, which was already answered.

The thread started with you reporting under the "problems" section a misbehavior in V-Ray as

Max subdivs not matching the amount of passes

.
That was never the case, the reasons for your oversampling in the scene non-withstanding: that is the intended behaviour as subdivs != samples, or total passes.

You were given the reasons, and were told how to obtain what you claimed wasn't doable (exact number of camera and secondary rays.), and warned that going about an animation that way would *not* lead to a consistent noise level across frames, ever.

You proceeded to claim Corona does it like that, and has no issue, hence i replied asking for numbers, which you have not (but i'd encourage you to produce. maybe I am entirely wrong here. Only numbers can tell for sure.).
I think my grip of the thread is solid enough, myself.
Glad to hear you are satisfied with the replies.

Whatever setup reason (or bug, who knows!) the scene has will be found and solved, but the two things are not, and were never to begin with, related in any way.
When the shading and lighting are simple enough, the variance in samples is low enough that max subdivs and average samples per pixel match, as expected, if fortuitously so.

Me, I'd have posted it as a question in the tips and tricks section...

[LudvikKoutny]

Well, I think it's getting a bit more technical then it needs to. If you see a rendered sequence, and it looks good, then there is no issue. I would never need to feed it through some noise analyzer to determine variance per frame. If on the other hand I see, with my eyes, that amount of noise changes abruptly between different frames, then I do not need to measure variance mathematically in this case either. I know it looks bad, I know I can not send it to client, I need to fix it.

Now I would like to remind you that V-Ray does not have this problem. My point was that I, personally, did not encounter this problem in Corona either, as long as I was using pass limit only, without additional time limit, or noise level limit, which are other two render stop conditions Corona offers.

But again, I am thankful for all the replies, and I understand there are valid reasons for that. I just never realized that V-Ray's Max. Subdivs do not reflect pass count, so it was my mistake to post it as a problem, if it's not.

[LudvikKoutny]

Hi,

I would like to revisit this as it is still causing me problems.

So I have this scene where I have Min subdivs at 1, Max subdivs at 100, and noise threshold at 0.1, which is fairly high. The scene reaches final noise threshold at about 13th minute, but spent another 12 (!) minutes cleaning up noise in one highlight in strong depth of field.


In the end, it reached over 50 000 passes, out of which 49 800 were spent on just cleaning those few defocused highlights. Now it is certainly good that Vray can pick noise with such degree of adaptivity, but I think it's quite counter-intuitive that it does not respect Max subdivs value to such a degree. If I set Max subdivs to be 100 (10 000), I don't expect it to be exceeded by a factor of 5

If the value in new Vray does not represent maximum amount of samples per pixel, that's fine, but it would be good to update UI to be more explanatory of that.

[^Lele^]

Do you have vrayFur in the scene?
Is your MSR to 6?
If so, a 5:1 ratio for the rays which traced fur is precisely as expected (in the worst-case scenario where adaptivity was never met), as for known high-frequency-detail geo the engine automatically reduces MSR to leverage AA rays (which are the right way to go about those cases.).
If those blades of grass also had very high intensity pixels, V-Ray had to make sure the noise level for those got down enough, and failed to do so reaching 50k samples (as the noiseLevel Re shows with values close to 1.0).
The total number of maximum (ie. without adaptivity terminating early) per-pixel samples does *not* change, it's just distributed better.
Not in the manual, debated to nausea, no less, on these very forums: Forum search is your best friend.


You may question the engine, as it's your want and style.
I'd rather look into my own scene setups to see if it's me asking V-Ray to do the unreasonable (like too intense lights, on too reflective surfaces, at too steep an angle, f.e., wrong or obsessive shader setups, and so on. Whichever the case, your noiseLevel RE shows precisely what, why, where and how much.) , or V-Ray being unreasonable to my perfectly valid requests (like my scene is physically to scale for modelling, shading and lighting, and there is way too much time wasted *without* an apparent reason.).
For hey, Science, and the only thing you have any chance to know directly, and accurately enough, is your scene, while presuming about the engine is a very long shot in the dark, particularly when you share absolutely zero of what may be causing the issue (cfr. Science: Peer Review of Experimentally-grounded Data.) not allowing anything of what you say to be proven right, or wrong, and forcing the reader to take whatever impression (as that's what it is, above.) you decided to leave your post with on faith.