About noise in general

[pengo]

I hope to not hurt noone, I told what i told (wrote), surely because depends on my knowledge of vray (even if I use vray since beta 0.45). Often tweak Irrad Map settings to have not artifacts on little or very little details (like mouldings elements and similar), or clean GI from the little noise, that I prefer to switch GI method to BF+LC but doing so the rendertimes grow so much. So I hink from what I read until now that often we spend a lot of time in tweaking and maybe make it calculate since the scene is complete with BF+LC is preferable, not even in this case sometimes the rendertimes are very very unacceptable so the paramenters tweaking is needed and sometimes make the rendertimes resonable is not so simple. So the our questions to get a faster tweaking for the fact that is possible to see other render engines that seems really simple in settings like the vray's "competitors" (maxwell, fryrender etc etc) So this tweaking time results sometime very boring. Obviously I want to use vray and I want to continue to use it but if it's possible to know more to get faster results would be nice so the thread that even the best artists in cg community is following (BBB3Viz, Pixela, Guthrie, etc etc). Surely is a question of subjective sensibility of the vray user, reached after using and using Vray but anyway is not simple to clean GI. Sorry for my bad english

[vlado]

So I hink from what I read until now that often we spend a lot of time in tweaking and maybe make it calculate since the scene is complete with BF+LC is preferable, not even in this case sometimes the rendertimes are very very unacceptable

Can you get me a scene where using brute force + light cache gives you unacceptable render times, just to make sure we are on the same page I've seen many scenes where the GI itself is not the slowest part, so it would be useful to see what's going on.

Best regards,
Vlado

[joconnell]

One major thing is that IRmap is quite separate from the final pixel sampling so your other settings have less of an effect on it - brute force on the other hand is similar to any other vray light in the scene, and it's subdivs are heavily controlled by the DMC and your aa.

[pengo]

to Vlado
I was referring to scenes of my works, (as many others users did here) interiors essentially with a lot of lights and a lot of glass objects and glossy reflections and refractions. Sorry if my english make me misunderstood. I asked an help about one of them some time ago, was asked to me to send the file, I sent it to your staff and when i got scene back I got it's settings very simple and not realistic. So not so simple even for experts as peoples in your staff. If considering I have not one pc at work, in DR, but 5 or six when are all free from work. For sure U can feel attached for my post but was an ask about simple way to solve noise and to understand if there was newer version of universal settings. If I resulted in some way not nice in way to approach in conversation was not my intention, as I told my english is not good. And about the fact qmc is fast is not proper the truth especially if used in that kind of scene mentioned above and with subpixel mapping not switched as the "orthodox" usage of this parameter suggests.If I correct remember even Francesco Legrenzi make notice this fact too, or I must be a CG artist recognized by the CG community to become believable in what is expressed in my post? I'd add the fact that recently the vray users saw in the online manual, the universal settings adaptive amount parameter changed in 0,9 and no more 1,0 so what was told in this thread and many others in the forum are not pure attach to the vray quality or its creators, as the bugs about glass refractions i evidenciate in the past are only some things happen and found by users. And about what I expressed here, are my thoughts about the settings, I didn't told vray is surely bad than other rendering engine. I was asking if some parameters can be used differently, making understand that probably I still not know a possible way to use some parameters to cut times as the example of subpixel samplig above meaned for. So the most responsability for noise or unacceptable rendertimes was given to me and my knowledge making understand that I suppose there is a way to cut rendertimes with better or different settings, but I don't know wich they are and I was asking if there are wich are. Of course I don't pretend u solve all the questions of each user can ask u can't for sure noone could and I recognize your availability to do it anyway, but seems that not only me asked about the universal settings if is changed and seems (to me I can tell) It's changed.

[windowlicker]

Not exactly sure I follow you there Pengo, but I can only tell from my own experience, and that is learning all of (or atleast most of) vray FOR YOUR OWN NEEDS is priority number one. I've never used the universal settings, since it has never been faster (quality considered) on any of the attempts I've made - from small scenes to very large ones. I've been using IM/LC since... waaay back, and still I think it produces the best quality in the least amount of time. But you gotta know what to tweak and when.

I'd love it if vray had the option to be able to analyze a scene - geometry, lights, materials - and be able to optimize everything to a global quality (noise) setting. That'd save me a lot of time in the rendering process. The actual time the processors need to render the image is an important thing, but the time it takes to manually optimize a scene shouldn't be ignored.

[pengo]

Maybe is really hard to me to explian the concept. Is impossible the software autoregulates itself evaluating the scene to render. I never would ask that excessive request to a rendering engine, because it seems to me excessive, I don't think is possible to get something similar. Maybe a gene could help about it. What I was talking about was the settings useful to avoid long tweaking. I think to know how to tweak Irrad Map I use vray since a lot and doing architectural interiors, exteriors, and design objects in studio lighting setup and so on. But sometimes to get clean results and advanced lighting i'm pushed to use BF+LC obviously not ever it happens. And using this solution with universal settings that is not the best in therms of optimization I agree with u windowlicker but is the fastest way to tweak settings available. Of course I tryed many solutions, and i found that exist faster way to get renderings with more optimized settings and as some other user explained before, everyone found his personal settings to face the long rendertimes this combined GI method produces. Doing tests me and many others discovered a lot faster rendering times and cleaner using clr threshold (lowering it from 0,01 to 0,001 in extereme cases) even untouching for example noise threshold value (leaving it at 0,01), so yet for this reason jumped at eye that maybe with the latest version some settings and values working in very different way, or if is better to explain me this thing confuses a little on how vray works, watching that even reducing the noise threshold was not possible to get cleanest results respect the default value, having at the same time great difference in therms of rendertimes. So the question escape not only from me, was is the universal settings changed or should it be changed? Avoiding for example the usage of noise threshold? And if it is useful to avoid its tweaking there is something other parameters to avoid that could not influence quality of the image and at the same time could make grow rendertimes? I made an example even menthioning the subpixel mapping that switched on makes appear the green warning message in the vray window message. It should be switched off for many reasons that reducing the quality of some kind of rendered images like with trees leaves with sunlight behind them but for the most cases is better to switch it on to get a lot more fast renderings respect switched off. So the question was and I repead from what I read not only from me if there are some other values that is better to leave as they are or not starting from the universal settings concept to avoid long rendertimes.Even store to direct light. Of course everything could and chould be relative to the scene to be rendered but even some parameters seems not works as in the past using to say a method that a lot of users use to give light to thei interiors, the domelight with hdr or exr maps in it with its gamma value lowered to get sharp shadows etcetc etc etc I was not asking clever software or miracles or offending or reducing the capabilities of vray . I still use it. Me, watching other previous pages of this thread, I'd say not only me found this noisy renderings strange respect the 1.5 version and tweaking settings as it would gave results to eliminate noise, discovered something I will not expected from the standard (how to say) vray behaviour and the opinions gone forward about the settings would have been better or not to tweak, and was asked the Vlado opinion, but maybe I wrote badly until now and everyone knows vray at its best and get always the better GI in low rendering times except me. Sorry the disturb if so

[vlado]

Wow that's one long paragraph I was just asking for a scene It will help me to understand what kind of scenes you have and where most of the render time is being spent; it will also show me what settings you have tried to use and maybe give you some suggestions on how to improve them. This, and scenes from other people are generally very helpful to see what people are doing.

The universal method has not changed at all; it works fine for me to reduce all noise in the image. But, again, there might be something else going on in your scenes, which is why I wanted an example.

Best regards,
Vlado

[pengo]

Vlado, sincerely I Thank you for the interest. What i got is the same problems shown by BBB3Viz, Pixela, and so on. That micronoise, that often vraylight subds, glossy subds in materials and the tweaking by the noise threshold seems not have power to avoid, and often only tweaking the clr threshold until 0,002 makes the differences so there was some kind of comparison about the experiences to understand if it happens only to one person or if happens to others too. To discover if is some way to realize materials tweaking the maps by colorcorrection plugin or whatever cause could it be. If the light trought the curtains or trough the window glasses or what ever else. For now I thank u again I have not a scene ATM or time to keep an old scene and retests on it the parameters seemed didn't give the results ever gave. As soon I'll can I'll try to demonstrate what I wrote as many others did in the previous pages

[Guest]

Hey Pixela, not sure if this has come up already in the thread but this
http://www.interstation3d.com/tutori...yfing_dmc.html
really cleared up some things for me on how anti aliasing is handled in vray.

Following the ideas in the above link would confirm what joconnell said that with settings like
( BF Subdivisions are 50, Material Reflection Subdivisons are 60! AA is min2, max50 with threshold 0.002)

your materials at 60 subdivisions are being divided by your max subdivisions of 50 or 60/50=1.2 subdivs rounded down to 1 subdiv for your materials so in that example you would still be handling all of your noise with AA and material subdivs are not really contributing.

BF Subdivisions are 50, Material Reflection Subdivisons 400!! AA is min2, max50 with threshold 0.002

as this would give you 400/50 = 8 subdivs

Just to go back to this comment by pg1. What I cant seem to decide upon and where I get confused on what is the better solution, correct me if I am wrong but this is how I understand it:

60 glossy material subdivs divided by 50 max subdivs in the adaptive DMC sampler is rounded down to 1 subdivision for that material. But it means that the adaptive DMC sampler now has a possible 2500 samples (50x50) to clear up that glossy reflection. So the adaptive DMC sampler does most of the work.

Or

60 glossy material subdivisions divided by 5 max subdivs in the adaptive DMC sampler equals 12 subdivisions for that glossy material which is 144 samples (12x12). But now only a maximum of 25 (5x5) samples for the adaptive DMC sampler. So the material subdivisons do most of the work before the adaptive DMC sampler takes over.

I guess its a toss up between locally sourcing samples or globally.

Then I think what the DMC sampler does globally across all subdivisions is work out how adaptive or not adaptive these calculations are and whether it needs to create new samples or use old ones.

[joconnell]

What I reckon at the minute is that if you've got a light that's causing loads of grain, you can try cleaning it up with AA but you're probably going to get a faster result just solving the root cause of the noise and turning up that light's subdivs. If you have really noisy results coming out of your light and material samples, that gets passed back to AA to do it's final checks, and it forces AA to try and clean up the mess. You end up with more loops of AA but as you've written above, the higher your aa goes, the more it divides your light and material samples, so the cleaning up of the lights and materials get less and less effective - they're almost being undermined by the AA sampler.

[Guest]

My next thought would be how do we know how many samples/subdivisions something needs? For example if a glossy material has a value of 0.7 how many samples would it need? Like wise if a light has a multiplier of 30 how many samples would it need. These are things I would love to know exactly but I don't have the time. So for the moment the answer is 32 and I keep the AA low.

[joconnell]

Right. I didn't want to write anything until I was total certain of whether my approach works or not, but I reckon the sample rate is the key to it. AA messes with everything else, so you've got to find what AA you need first, and that's via the sample rate element and getting a nice balance with only red pixels on the edges, and fine texture detail. Once you've got your AA set, then start looking at your light and material samples. If you're constantly changing both, you're chasing your tail. 32 samples could be okay on materials but some will need more, some will need less. What could be another good guide is the mtlglossiness element, so you can look at your render and the grey values tie in with how glossy a material is, the darker, the more blurry. What you can do is swap between your reflection pass and the MTL glossiness pass, and see what materials are noisy and clean. If all of your 0.6 glossy materials are clean, then 32 is likely okay, if not, up the samples.

[peterguthrie]

Can you explain what one should be looking for in the sample rate element a bit more? Would love to be a bit clearer in my mind about that.

Does red pixels mean its using the max value? With universal settings, you typically just get blue values, which I assume means its never really needing to go all the way to the max value (100)

[joconnell]

Yep - sample rate is like a heat map with red being whatever your max aa was, and blue being your min. I think to get a good level of aa, you want blue in the flat boring parts, and red along the edges of geometric detail, complex fine detail in textures and then likely on highlights / burn out areas.

And yep with the universal method, you're giving vray the option of going up as high as 100 max, but it'll likely never do that so you'll only ever get blue.

[Guest]

I asked Chaos Group yesterday as I had a render that had red, green and blue colours in the VRaySampleRate render element. They replied with blue areas show less samples, and red with the most samples. Green is somewhere in between. When you get various shades of say blue I think it means the samples are very well distributed across the scene and red means you are over sampling too much. So best not to see red at all (could be wrong though).