About noise in general

[cecofuli]

Hi Vlado,
ohh, yes, sorry... I forgot to send you the files!
Now I see, BF= 256 subdivs and VRaySun 256 Subdivs. I think for some old test :/
The rendering time is less (obviously) but the difference in brightness is the same.
PS: your PM box is full

[Morbid Angel]

Actually I sent vlado your old scene with the chairs...

[cecofuli]

Ahh, ok!!! Vlado, which weird parameter?

[pengo]

I would be interested in the weird parameters that should be avoided, if could be known I think would be useful not only for me. If this thread reached 31 pages maybe I'm not the only interested in what to avoid or to choice to make the renders faster and better in quality about the noise.
Thanks in advance.

[Guest]

I have been reading this thread as it goes along and doing my best in taking in all the various workflows. It would be great to have some sort of summary of all the different settings people have and how they compare

[joconnell]

1 / 100 is a quick way to get predictable results but not with the best times.
Dmitry's 2/16 guarantees totally clean renders but at the expense of potentially nasty times.
1 / whatever is appropriate has much more tweaking time but the best combination of quality and time.

[Guest]

I never use BF as I am not confident in getting the results I want. My setup has always been:

3ds Max Preferences:
Gamma 2.2
affect colour selectors and affect material editor ticked
bitmap input gamma 2.2
bitmap output gamma 1.0

V-Ray frame buffer enabled

Adaptive DMC
Mitchell-Netravali filter

Adaptive DMC:
Min 1 Max 11
Use DMC sampler thresh unticked
Clr thresh 0.007

Colour Mapping:
Reinhard
Burn value 0.9
Sub-pixel mapping and clamp output on
Dont affect colours off
Gamma 2.2

Primary Bounce: Irradiance map
Preset medium
HSph. subdivs 70
Interp. samples 30

Secondary Bounce: Light Cache
Subdivs 1500
Light cache glossys on
Retrace threshold on at 1.0

I tweak all lights and materials manually but I stick to a typical range. Material subdivs range between 18 and 32 depending on glossiness value. Light subdivs 50 if HDR or not. DOF usually about 18 - 22 subdivs. If I haven't mentioned a setting then its at default. These settings may not be "correct" but they give the results I know work with no blow outs, splotches, noisy reflections or super bright pixels. I am just curious how this compares to others.

Render times for a typical interior at 3k range from 2 hour to 5 hours. Waiting 12 hours for a render isn't in my line of work, I couldn't wait that long. Oh and my whites are 235/235/235 and blacks are 1/1/1

[joconnell]

What DMC threshold are you using? Sub pixel mapping and clamp are definitely giving you an advantage!

[Guest]

Default 0.01, never touch it. All DMC sampler settings are at default.

[pengo]

I'd prefer to use Irrad map too for firt diffuse bounce, but often I get not a clean and defined results with little details, so I have to choose QMC+LC. I avoid some GI artifacts and I get more precise GI, but the times sometimes are really unacceptable. Yes is possible to tweak them but loosing a lot of time tweaking.

[joconnell]

That's exactly it - no matter what, money is the solution. Either spend more money on your own time tweaking settings to get what you want, or throw money at more render power to get things done quicker.

[pengo]

Of course it has sense Joconnell, but the question was if there is a method to have a GI solution faster with a newer Universal settings with the newer build, because seems that is not needed so much AA (min/max ratio value) to clean material and GI solution, instead the clr threshold does a lot in therms of works, I tweaked and i got faster results using 0,001 as clr threshold and (nois threshold at his default value 0,01) and only 1/24 in AA using obviously everything with their default subds. But to reach this results i spent a lot lot time tweaking and this fact, makes me think that my final (for now) solution is not the best in therms of tweaking. If someone ask to me why to use those extreme parameters, the reply is that is not simple to get clean GI with IM+LC with very little details (for me) and reducing the IM values crancks rendertime so much that is better to choose a secure GI method that avoid artifacts and gives better details, at the same or similar time respect the IM+LC. All I have told is obviously my modest opinion.

[m_hinks]

That's exactly it - no matter what, money is the solution. Either spend more money on your own time tweaking settings to get what you want, or throw money at more render power to get things done quicker.

Yep, thats it exactly. Personnelly, my time is better spent. Get a couple of render nodes, and leave them to it.

[vlado]

if there is a method to have a GI solution faster with a newer Universal settings with the newer build, because seems that is not needed so much AA (min/max ratio value) to clean material and GI solution

This is something that has been bothering me as well - I mean, what would be the best way to control the AA samples vs the samples for GI, reflections etc. The current method works, but it might be difficult to arrive at the optimal settings for a given scene. Perhaps some way to control the "ratio" of the AA samples compared to everything else might work better.

Best regards,
Vlado

[joconnell]

That's the conclusion I've come to vlado, that the AA sampler bullies all the other samplers and thinks it can clean up everything else, but it doesn't make sense in most cases especially where more light or material samples could have solved the problem easier. The sample rate element and the new mtl glossiness elements are great helps for finding culprits.

With the early termination idea, in theory you could use ridiculously high samples on everything and they'd never get used, but I'm trying to think would it be useful to have some scene stats where you'd be told what level of subdivs a material or light actually got up to so you could make better guesses what needs more or could get away with less. I think you mentioned before that other samplerate style elements were very memory heavy (and you can pretty much use the render passes anyway) but maybe some other max samples stats even at a text level might be useful?