Excessive amount of noise

V-Ray for 3ds Max V-Ray for 3ds Max :: Issues
,
1

Hello,

We are currently working on a disco club, and spent last day trying to clean up noise in image changing various vray settings. Nothing seems to help.
Scene is fairly complex in terms of lights - uses predominantly mesh lights and several vray light materials. It has several vray lights as well as maing source..
Initially we tried with Irradiance map +Light cache, but after having endless calculations, switched to BF+LC. After 8 hrs of rendering VrayLight passes of all lights are super super noisy in dark areas. They look like big chunks, which is super weird. We tried resetting vray settings, merging into a new scene - same. IM+LC seems to be almost same speed as BF+LC in Progressive mode.
This is not the first time we work on such kind of project and to be honest we did not have any issues several months ago with another and earlier version of Vray 3.2.

You can see the noise in top right corner, almost everywhere there is a mesh light. Image took 13+ hr on single 4930k, whereas similar scene with twice more vray lights (Vray 3.2) took no more than 5 hr.

On a recently completed interior scene we had similar problem- huge amount of noise in white material which was behind glass. Fix it by increasing Min Shading Rate to a big number, but render time increased almost twice.

Has someone else experienced similar issues, or we should delve into the scene more? Unfortunately deadline is knocking on the door. :slight_smile:

Thanks in advance!

Paradox11.jpg
Paradox11.jpg1463×823 440 KB

2

get rid of mesh lights imo they only cause issues. Replace with vray lights, shaped as planes, spheres, discs etc.

3

agree i think there is something wrong with mesh lights - replace with normal lights and it will be much faster.
i use them very sparingly

4

Thank You all for replies.
We managed to deliver images on time, but turned off the mesh light and used only vray light material only applied to these curved shapes. Noise was still visible after 15 hr of rendering though. Luckily time was on our side.
Later on we did some tests on certain regions and found out that raising MSR to a much higher number like 64, or 128 gave us cleaner image within same amount of time. Especially in GI and Reflections, despite making Light pass a bit noisier.
The drawback was that instead of fast 3 min passes(MSR set to 6 ) we had 30 min per pass(MSR set to 128) in Pogressive rendering mode.
Still not getting really familiar with how to optimize renders only by using MSR, although being long-time Vray user. :slight_smile:

5

The light material has a bug in the current release.
There’s another topic around where someone was having issues with difficult to clean noise, and Vlado tracked it down to something in the lightmtl.
If you could share (to my chaosgroup email, ofc) a scene, we’d be grateful, anyway.
Mesh lights don’t currently have any reported issue, and should work with performance much similar to standard light fixtures (for the same fixture shape and size, of course.).

6

tangent incoming..
honestly i don`t know if it`s particular scenes being awkward but in vrays attempt to become `simpler` it seems to have become much more complicated. trying to get clean images seems to need a phd now where it used to just be a better quality vs time scenario. now you have to juggle the shading rate/noise threshold/aa subdivs etc to see what works per scene. for an example of this you only have to read any thread explaining about how any of these settings work, math all over the shop. i never needed anything but adaptive 1/4 0.005 before, now it`s a bit of a guess what settings work. now i`m probably doing things wrong i totally accept that however vray used to work with me whereas it seems to now only want to work one way (which obviously i`m not be in tune with yet). i`ve seen the `cheat sheet` thread but all the kerfuffle you have to go through with render passes to show samples etc. and again adjust on a `per scene` basis just isn`t workable. in an animation/production environment (at least ours) clients give us no time for any of that. i`m considering downgrading for now to see if it works better for me presuming we still can ? (or whether this is just rose-tinted spectacles). don`t get me wrong, i want vray to be simpler, but i have to be working, not constantly playing with settings. the impression i get is that maybe vray has been improved a lot for still images etc. animation not so much.
anyway don`t want to seem like too much of a rant, scenes misbehaving causing grief that seemed to render fine before and now not are poking me with sticks.
sorry for the diversion, return to your scheduled programming.
anthonyh

7

It would help if you showed us what the problematic scenes are.
The decision to go the way we did was taken to avoid users’ mistakes (and god knows there were a LOT of those.) which led to over and undersampling.
The method is by far the strongest V-Ray ever had to get to a clean result regardless of the scene in question.
That users decide we had it all wrong, as we clearly have no clue on how to use V-Ray, and dive in to break what defaults were set is unavoidable, to a degree.
We can only give you tools, and methods, and write time and again the exact same words, but we can’t force anyone to read, or do, whatever it is we think best.
Bugs and mistakes do and will always happen, and that is why we take great care in looking at users’ scenes and their setups, but you’d be surprised to discover most of the times the pain is well self-inflicted: break defaults, tweak this because i’ve always done it this way, and do that because i know better. THEN it breaks.

It’s not far removed from a car dealer selling you a car, and telling you to use the accelerator to move forward, when the clutch is fully released.
If you think you ought to give that clutch a burn-in because your 10 years old old car needed it, and then burn the clutch, how’s that the fault of the dealer?
If, however, you tried PRECISELY what you were told, and THAT doesn’t work, then yes, the problem’s with the dealer, and you can return the car for checks.

Namely, in your post above, i do not see the issue at all: get the render settings at their default, render (a crop.), if it’s noisy, lower NT, if it stays noisy, raise max AA.
Render time will of course vary, and that has ALWAYS been the case.
How did you set up your scenes before, pray tell?
If adding a SR RE is a deal breaker to you, i can’t help you there, besides adding it automatically, which has been considered.
Then again, you’d have to go all the way to the drop-down menu at the top of the VFB and change channel.
Which likely would be “keruffly” as well, given your tone above.

Sorry if i sound slightly miffed: i am.
Send me a scene, and then we can have a debate on what is wrong with V-Ray and what is wrong with the User’s setup which went against the suggested workflow.
Unsubstantiated, wild, catch-all claims get my blood boiling, but that’s no news, hey.

8

Its currently the easiest it’s ever been. With the denoiser making it even easier.
Just use the default sampling and press render. If it really is still slow/noisy then at least you have a solid base to tweak from.

9

hi again,
no need to panic, you`re not under attack, there`s no `tone` and no wild claims, and i suggested many times it`s the way *i`m* working with vray. internet message are impersonal and easy to see issues where there aren`t any. although you`re analogy of a car dealer telling you what you `should do` and it`s your fault if you don`t is very apt, it was my point too. vray seemed more flexible in the past, at least to *me*(i`ll add that qualifier from now on!), i could mess with the settings `at my peril` and happily live with the results. vray 3 seems more exact in the way i *should* work. my view on this may change once it clicks, who knows.
you`re right in that leaving default settings is the best way and that messing with settings can get you into trouble although i never had any major problem before, there were always little tweaks to help, global subdivs/noise threshold and certain motion blur tweaks etc. that worked better than the defaults nothing ground breaking, they just don`t seem to work any more for *me*. having gotten used to one way of working it`s always a tear to have to change my ways. having barely used brute force/lc before it seems that`s the way it`s going now, not a bad thing since it`s much quicker although not as quick IR which i used mostly. i`m not yet convinced it will be always quicker versus what i could live with before but we`ll see.
as you say, the tools are there but i don`t often have the time to use them all *properly* to achieve deadlines, it`s really just that. denoiser, i`d love to use but it`s another step to add and it all takes time. factor in rerenders galore because of painful clients and every step counts. give me loads of time for every job and i`ll gladly do everything by the book.
once i get more free time i`ll gladly send a troubled scene.
until then shouldn`t be rambling on again since i genuinely don`t want to raise anyone`s blood pressure.
anthonyh

10

No worries about my blood pressure, it’s generally kept all too low anyways.
The issues you are facing are the issues we are facing with many old-time users.

Namely, the affection to tweaking which was a requirement for quick renders before 3.3 (everything had decent defaults, on its own, but may or may not have played well in concert with other things).
As of 3.3, DoFed and MoBlurred scenes even tend to render quicker than plain scenes, for a given noise threshold, for example.
Which makes perfect sense, as both effects blur the image, so they lower the local contrast, so the sampler ADAPTS, allowing the sequence to clean faster.
That was impossible (not really impossible, but VERY convoluted to achieve. I know, as i made a movie and a tv series with similar approaches to sampling) before 3.3.

So in a way you are right: there are differences between the defaults of the old V-Ray, and the defaults of the new one, and on how it all works together.
But while you could achieve precisely the behaviour of the old V-Ray with 3.3+, you would be seriously hard pressed to achieve a similar quality in a given time with versions which came before 3.3.

I wrote it in the cheatsheet post, a few times: “Refrain from tweaking”.
At least as a first approach to something entirely new, that looks to me like very good application of game theory.
That there may be corner cases in which the new sampler finds it harder than the old approach, we can all agree on.
That each and every user should find those corner cases to be their daily experience, when NO TWEAKING has been done to the settings, i still have to find proof of.
And that’s why i keep asking for scenes, and so far i had a grand total of zero cases which had the settings at their defaults, a year and a bit from the cheatsheet post (including the scene which started the thread, non-withstanding the issue with the lightMtl.).

I’m not panicking, i’m not replying here because i feel i am under attack: it’s part of my job, and i am genuinely trying to help, which isn’t identical to laying down and get two years of hard work of a committed team lambasted and diminished without the shred of a proof about the claims.
That you should read it as a reaction to some personal insecurity of mine is cute, but quite wide of the mark: take a look at the new website, and at the client’s work on display there, and you’ll see why that isn’t the case.

TL, DR : I can help fixing problems, i can’t fix unsubstantiated perception.

11

hi,
even when i blame myself i`m still the bad guy. i`ll cut my losses.
thanks
anthonyh
that sounded curt, sorry.

12

I don’t see who painted anyone as the bad guy.
I tried explaining precisely, factually, objectively, what the issues are.
Including that of the evanescent perception, when any change happens (it’s a very common thing for every living being, really, including me.).

If it sounded harsh, i am sorry, it wasn’t so much personal, as generic.
I shall wait for the problematic scenes, for whenever you’ll find the time to send me one, so to at least give you back a fix, or a quicker render, in return for the misunderstanding.

13

hi lele,
well i`m working on one scene right now. just trying the `cheat sheet` method. it`s working to a point. using BF it never truly seems to be what i`d call `clean` like what you`d get with the IRmap although trade off is less flicker for animations ? maybe that`s it, i`m use to using IR not BF so it`s naturally more noisy at the start. it`s getting better but i`m not sure i`ll get the IR equivalent clean without heavier render times, the rub is that IR medium/high animation in this scene is not working at all well.
i do understand i`m supposed to use vray 3 differently. i`m trying but probably been resorting to old habits.
i also apologise, messages are easily read as being angry rather than just jotted down without thought about how it`s read the other end. as you say these were only my perceptions of how i`ve used vray in the past. i`m also sure you`ll have a field day with any scene i send, no gamma, no vray frame buffer. i`m gonna get clobbered. it`ll be a slightly truncated scene as it`s client cad. who to send it to ? can`t be public i`m afraid.
anthonyh

14

Eh, yes, the defaults for primary GI are now BF, and again, while counter-intuitive at first, it’s by far the best general approach (it will not flicker, or blurb around, regardless of what’s thrown at it).
I have written somewhere else (but can’t recall where.) on the forums on the benefits of moving away from IRMap, and on how that will require some readjustment of expectations: it may render longer - often not by that much - but it will never be an issue in and of itself - unlike the IRMap -, and it will require zero setup -again, unlike the IRMap - .
It’s not that the IRMap is bad, it’s just that the times have moved on since its inception, and heyday.
And what i mean by that is that 10 years ago no one dreamt of the average detail a modern scene contains: stuff was broadly composed of flat, even surfaces, and we were all very happy we got some light to shine into that corner.
Today, people add fur to carpets by default, 4k textures on the minutest of detail, and i do not see why they shouldn’t.
However, that negates most of the benefits the interpolated approaches provide, by virtue of the fact very little can be interpolated when the detail frequency approximates, or exceeds, that of a pixel (say, fur in the distance.).
There are cases where for a still, or a camera animation sequence, that will work well, and quickly, provided one doesn’t really look to close at what the lighting is doing, but the moment one has animated lights and/or geo, the requirements for the IRMap calculation will approach, or exceed, those needed by the BF calculations (that aforementioned fur on a distant carpet, f.e.).

No worries about the misunderstanding, and again sorry for the harsh tone: hit me at emanuele.lecchi at the chaosgroup domain with a scene, and a description of what the issues are, and how you’d like it to behave, and i’ll see if i can be of any help (words are cheap. mine all the more so. :wink: )

15

hi lele,
i finally had a gap to send you a scene. it`s a bit *gutted* due to client cad. basically IR/LC is 3mins a frame. BF/LC is up to 10-30mins depending how clean you want it. i`d appreciate your thoughts. i tend not to work *by the book*(gamma 2.2 etc) so maybe it`s something obvious ? i`ve also got another problem scene with lots of white trees(just white vray material) exterior shot and under the tree canopy is a haven for noise but i don`t want to overload you. sent via external upload.
thanks
anthonyh

16

Well, why do you not use the V-Ray VFB, and decide to work with compressed dark tones, and no linear relationship between what you input and what comes out of it (lights, shaders, any visible color, and product thereof)?
The scene you provide has a medium IRMap, which renders a crop, here, in 20 seconds, the SR RE entirely blue.
It’s blue because you aren’t under LWF, and the values in that image are tiny, in absolute value.
So tiny, even their percentage variation is negligible, well below the threshold you set in the scene.
In fact, the IRMap skips over shadows quite happily, resulting flat and blobby, and misses out on chunks of rims and fine mechanical details.

So, you are right, there is no time-comparing this scenario with a BF+LC as that’ll sample multiple times per pixel, instead of a few every a few other pixels, but in your case will still look very noisy, because of the non-LWF setup, as the sampler settings are thrown off-board: you will need four to five decimal places to have any hope of being precise with the way you drive the sampler through Noise Threshold.
Further to this, what the sampler sees, and what you see in the VFB (Max or V-Ray’s) are not the same thing, as by not going LWF you apply a gamma to the final output by means of the display device, of which V-Ray can know nothing of (that’s why you’d set up the 2.2 inverse gamma, btw.), which exacerbates noise as it stretches the pixel values back, after V-Ray’s done sampling them (ie. you contrast it).

Resetting the render settings, still without gamma, and switching to bf/lc produces, as you said, slower renders, depending on the amount of visible noise one wants, but with a VASTLY superior GI detail, down to the finest crevices of the complex geo you have, which you could never know existed from the IRMap setup.

However, being outside of linear workflow, there is little that i can do to help you: it’s not a fancy thing, it’s kinda been a requirement for any CG displayed on 8bit monitors for over a decade, and V-Ray needs it to operate at its best, as it has had for as long as it could (well before the switch to 3.3, and before 3.3 even the more so.).
Try aligning yourself to LWF, leave the max and v-ray defaults alone, unless you have a very specific need to change a thing, and then we can talk about performance: it’s difficult to fathom what is it you expect as quality and looks under this setup, for me, and i fail to see -besides needing a much lower NT with 3.3+ because of the percentage instead of absolute value- how 3.3 broke anything in respect to your workflow with IRMap and LC.
You’re missing out a LOT more on not being under LWF, and not using the V-Ray VFB.

17

hi lele,
well i did say i don`t always use vray *properly*. w
generally because clients love to change stuff down to the wire and i need to be able to re-render large chunks of animation very quickly, i`d render everything full quality if i had the time. i often use IRmap because generally most clients wouldn`t notice the quality difference and deadlines always loom. it`s why i`m in the office at the weekend re-rendering everything, again.

VFB i don`t use because i do all post in AfterEffects, is there any advantage of using the VFB before saving ? i know i can tweak exposure etc but any grading i do outside 3dmax.

i`ve been trying LWF with that scene i sent and it does render cleaner/quicker than in gamma 1.0 by a decent margin(i really didn`t know LWF would affect render speed so much) so i`ll use that from now on. i `m having to go very low 0.001 or even lower to get it `clean`though, is that normal ? it renders 10-12mins per frame at fairly decent quality which is pretty good and a little quicker with a bit more noise so it`s getting closer.

thanks for the explanation and taking the time to have a look at the scene.
anthonyh

18

It’s my fault, i needed to get an LWF sticky up a while ago, and i fiddled instead.

The VFB, while perhaps not perfect, it’s a huge tool to help you light and render.
The mouse follow/mouse paint feature alone would be worth it, but the channels/RE list, the render region, the connection to PDPlayer, the tiled/scanline multi-layer EXR saving, the ability to save deeps, and so on and so forth, can only be had through the VFB, as it circumvents Max’s own limitations.

That aside, i believe you will benefit from lowering the overall Albedo (read: total reflectance, improperly “brightness”) of your shaders by quite a bit (say, start from half the Value, as sum of diffuse and reflectivity, and move on from there.).
This will return you to a more accurate light bouncing behaviour, and cure some of the washed-out look moving to LWF seems to give at first: it’s the surfaces not extinguishing incoming light properly, now visible as the monitor gamma has been corrected for.
As a bonus, because light falls off more closely as it does in real life, and much quicker than currently, you will get fewer instances of fireflies, and generally (much) quicker renders to a given cleanliness.

Notice you CAN and should use the IRMap if it floats your boat, i wouldn’t want to sound as if it’s a bad choice: in fact if you have scenes with plenty of flat surfaces, it’s got an upper hand on BF.
However, at least add Detail Enhancement to it, so to cure the detail breakup around the finer bits, without having to resort to BF across the frame: in your case you ought to be able to get the best of both worlds with little rendertime expense (don’t go putting hair all over the place now, though! ^^ ).

19

Wow, this thread took a bit different direction than it was supposed to.

Client is back, so the new deadline of 1-2 days for all images, comments and… even more lights. Typical scenario.
Thanks to Lele, we managed to cut rendering 2-3 times! We really appreciate his help!
Problem was due to merged Vray dirt extatex from another scene ( Vray 3.2) which obviously does not resets Subdiv settings. Turned out that Vray light Material is causing big trouble to this scene . This in conjunction to having pretty low Retrace threshold in LC, and not having Probabilistic lights turned on, gave the slow rendering time. Last one is something Lele adviced to always leave ON, although we have always had problems with it, even for far more complex scenes( more lights) and have had same fast speed as with Probabilistic lights Off. We also observed quite slow render times using newly Vray disk light, around 60-70%, but this could be only in current scene. We do not pretend to be masters, but use the tool, from quite a lot of time, and it drives us nuts, when instead of communicating with clients, deal with " what is wrong now , so it renders slow".
Reverting to earlier Vray version is not an option for us.
For good or for bad, over the past years adopted the " all 3d approach", which proved to be working for having short deadlines, small budget, and ongoing changes.Community become bigger- it is easy to find 3d models and put it in a scene rather than photosop a view which eventually might “can you move the camera back a bit”. Since being based in Eastern Europe is “disadvantage” in 3d world, although most of You might argue, we try to bet on speed(modeling included) and somehow adequate quality. “Can you fake it” comment from client is the least viable option for us, so we just want to put everything there, models, lights, proxies, test default settings for max of 1 hr and just render it. 10 hrs must be sufficient nowadays to render even the most complex scenes. (all render times given a per single i7 4930k) .Remember doing quite complex scenes in the past with Vray 2.4, even 1.5 , and having old, single Q6600 - even then have never had 10+ Hrs rendering on a commercial render. This also speaks that we( team here), as artist have not moved forward that much - having latest version, newer options, several times less render time - and still not getting anywhere (irony).
Again- we are small studio and can not afford buying Xeons, render farms, etc. It turns out that in some terms we are wasting more time on adjusting Vray than even focusing on postproduction- unacceptable! By adjusting mostly mean setting some of the settings to Advanced mode and check AA/GI/Shadows/Glossy appearance. Most of the Vray settings work great, out of the box, which we should be thankful to Chaos Group.
Returning back to this very scene, which got a bit more complex- around 30 more small Vray sphere lights( all instances). With already optimized settings, scene got slowly again. Jus reset Vray settings again, not getting even closer to render times after the optimization. Amount of passes, which clean noise threshold even got higher ( quite fast though 3-4 times). Here comes another question - what is the point to have faster passes in Progressive mode, when instead of 13 passes(used to be in “Slow” mode 16+hrs, to clean certain Noise Threshold we have 40+ ! So again, something forces Vray to do things in another way, but what is it - who knows.
Knowing nothing in virtual world is linear, and looking at GPUs - which are almost there( 1+1, twice the speed). So again Hardware is ahead of Software in terms of predictability, although Moore’s law is considered to had slowed down considerably over the past decade.

So… apart from the faster rendering, which is undeniable success for most of the scenes, we got surprised that there are known bugs with Vray Light materials, merging Extratex passes from previous versions and maybe others, which more or less influence speed. So the logicaly questions is - Should we update Vray when newer official release of Vray is issued or wait?Kind of of fed up of what is fixed, compatible, and still no single world- known bug with this and that. At least the community should be aware of this, since we do take our time of write down, share scenes, discuss give insights. Maybe a small thread with all noticed issues with latest official version might be a good approach, although not marketing tool at all.
As artist, we know the tools necessary to deliver images, but should we all have PhD and delve into the basics of render engine again and again?! Used to be numbers, then dividing some of these, then checking light pass, reflection pass and GI, then changing BF as main GI (which is great), and now it is just miscellany of everything (something for everyone), mostly hiding behind several checks.

Don’t get me wrong - we just want to leave Vray do the job, within reasonable amount of time, as it proved in 99% of the cases. We would be happy to have problems with 1 out of 100 projects, or scenes, but it is becoming more unpredictable recently. This project is a great example of why we should stick to the “working versions”, rather than updating to newer ones. You never know if next project/scene would be this single out of hundred, which does not work.
Exterior images- fine, no worries there Sun, Sky - boom, atmospheric effects - all great, but interiors, custom ones - with mix of different light sources and Vray light materials, hint of glossy - so much pain :frowning:

20

Little update - we got rid of a big object that had Vray light material applied onto it. The scene rendered within reasonable amount of time, this object seemed in conjunction with this material seemed to slow down things a lot.
Since we have another camera, positioned in same space, we just switched to it with smiles on our faces, until… 2 hrs no trace of noise threshold decreasing. Unfortunately we have deadline which we will miss because of this problem, since we are unable to localize the problem.
Tried with Buckets - 1/24, with Threshold at 0,001 - results - Sample rate redder than tomato. 1/100 - Threshold increased to 0,003 - no trance of a blue colored pixels. -All light off- same. Strange thing is that even with all lights off and Vray light materials with Swtiched Off Direct Illumination - these materials still cast shadows.
What is wrong ?