Antialiasing 1.09 to 1.46. where's random and threshold gone

[DP]

Ok we nailed down some pretty good Simple 2 level settings in 1.09.x for use in max 5...now we've jumped to Max 7 and 1.46.x and it seems Apaptive QMC (same thing) doesn't have Random samples for AA or threshold anymore...

Now, do we have to do a whole load more AA tests using Adaptive subdivision or is there a way to switch them on?

Thanks...

[arobbert]

I always used simple 2 level in 1.09. I spent a lot of time working with adaptive QMC and adaptive subdivision. Adaptive QMC was always a little noisy in the glossy reflections. No matter how much I tuned up the settings and QMC I couldn’t get satisfactory results. I decided Adaptive subdivision was the best option but still think Simple 2 level is the best speed/quality for my purposes. I asked about why it was removed about a month ago and no one responded. I don’t know why they got rid of the multi pass option. Adaptive QMC is not a replacement for Simple 2 level as far as I can tell.

Alexie

[vlado]

The Multipass option is always ON in 1.46 since it made little sense to turn it off.

The Rand option is always ON if the number of subdivs is more than 1.

The Threshold value is taken from the QMC Sampler's Noise threshold. If you do not reduce the default value, you will always get to the point where more subdivs in the material/light etc do not reduce noise anymore.

The reason why the AA threshold and the QMC Sampler's Noise thershold are linked is because they are part of the same QMC sampling system; if they are different, it is very easy to mix them up in an inefficient way.

Best regards,
Vlado

[arobbert]

I've tried reducing the QMC threshold down to .001 with 12 min samples and min 1, max 4 for adap QMC sub aa and still pick up noise in reflections. Am I doing something wrong? I'm curious why did you get rid of simple 2 level?

[vlado]

If you can post a simple scene that you are having problems with, I'll take a look and explain how to reduce the grain. Otherwise I can post an example too, but it will be a teapot on a plane

Best regards,
Vlado

[DP]

thanks vlado, we'll see what shakes loose

[arobbert]

Thanks Vlado. Could you post your scene in MAX 4 or even sample images with settings. A teapot on a plane would simulate my problem. It's in the plane that I'm picking up the noise. Any chance of getting simple 2 level back or do you feel this new one is a better option? Thank you for your reply!

Alexie

[vlado]

Here are a few examples:

Default values:
8 subdivs for the material, qmc sampler: 0.85 amount, 8 min samles, 0.005 noise threshold, 1.0 global subdivs mult; adaptive QMC AA with 1/4 subdivs; render time is 12 seconds:


Here I have increased the min samples to 16 and reduced the noise threshold to 0.003; render time is 14 seconds:


Here I have also increased the global subdivs multiplier to 16.0 (the noise threshold and min samples are still 0.003 and 16 respectively); render time is 40 seconds:


Here I have reduced the noise threshold to 0.001 (the other settings remain the same); render time is 93 seconds:


As you see, there is very little grain in the last image; those settings work very well in many cases. If you need faster render times, you can increase the noise threshold.

Best regards,
Vlado

[arobbert]

Thanks Vlado,
It looks like you do all of your adjusing at the QMC sampler level and not at the adaptive QMC aa level. I hadn't tried increasing the subdivs multiplier I'll give that a try. I assume that increasing these QMC settings will also increase GI calc time. So should I reduce the settings for GI calcs and increase them for rendering?

Thanks for taking the time to help me!

Alexie

[flipside]

Hi vlado,

How can you campare the global subdivs multiplier with the material subdivs?

are these two the same:
material subdivs=40
global=1

and

material subdivs=8
global=5

They don't give the same result in rendering. How should you do the math for this one?

kind regards,

wouter

[vlado]

I explained this once, but I'll repeat it here:

40 subdivs mean a maximum of 40x40 = 1600 samples.
8 subdivs mean a maximum of 8x8 = 64 samples.

So, to get the same result, you will need 1600/64 = 25 for the Global subdivs multiplier. It would probably have made more sense to square the Global subdivs multiplier internally in VRay, since in that case the math would have been somewhat easier...

Best regards,
Vlado

[flipside]

Yeah sorry vlado, I figured it out, it's the same as the hsph subidvs you explained once