Noise level limit - different pass

[Zray]

Hi. I have a strange issue with render passes. A scene with noise level limit =2.5, 2 my computers have different passes
Dual 2683v3: 25 passes
Single 2696v3: 30 passes. The quality of 2 images are little bit different as well, could anyone explain the reason?
Ps: my Crn is up to date. Thanks

[maru]

Is this exactly the same version of Corona on both PCs? Which version is this?
Is distributed rendering involved here, or just local rendering?

[TomG]

Quick note - you can add Noise Level to your render stamp using %pe.

The main part of the cause is noise level is evaluated every 5 passes by default, as that is when the adaptivity is recalculated (see the devel / debug settings). My thought is that noise level is never exactly the same each time you render, so one time at 25 passes the noise level calculated at 2.51%, so it did another 5 passes as it hadn't reached the 2.5 yet; another time it calculated at 2.49% at those 25, so it stopped. In general getting + / - "adaptivity recalculation passes" is quite normal, as far as I know - unless the developers say differently :)

[Zray]

Is this exactly the same version of Corona on both PCs? Which version is this?
Is distributed rendering involved here, or just local rendering?

the latest one, 1.7 hotfix2. I'm sure that there is no distributed render in this case. Do you need me add the noise level to the render stamp?

[maru]

The part about adding noise level to render stamp was just for your convenience - however it would also tell us what were the specific levels in your images.

I think Tom's theory about noise level being recalculated every 5 passes makes sense. There was just some tiny difference in the noise pattern (as far as I know this may happen on different hardware) and one image needed just 25 passes to reach the noise limit, and the other one needed 5 more passes.

[Zray]

The part about adding noise level to render stamp was just for your convenience - however it would also tell us what were the specific levels in your images.

I think Tom's theory about noise level being recalculated every 5 passes makes sense. There was just some tiny difference in the noise pattern (as far as I know this may happen on different hardware) and one image needed just 25 passes to reach the noise limit, and the other one needed 5 more passes.

Thanks, it is just a confusion as I thought a file has same passes every time, that doesn't cause any issue with quality.

[Frood]

It would be still interesting to know the exact noise levels of both tests. If the difference isn't really tiny this should not happen.


Good Luck

[Zray]

Here is some information for those who are interested. All computers installed Crn 1.7 hotfix 2
- With noise level = 2.5:
2696: 30 passes (I tested with another render node, it has the same result)
2683 x2: 25 passes, render time: 10m42, noise 2.49
- With passes = 30:
2696: no change
2683 x2: 30 passes, render time: 12m10, noise 2.25

[Frood]

Thanks, that means you get a noise level difference of 0.22 with the same scene+passes and by only rendering on a different cpu. With current daily I got only a deviation of 0.02 between some i7 and a E5-2670 x 2 (tested with 30 and 50 passes, 3 different cpu via Backburner, locked/unlocked sampling pattern).

It could of course always happen to reach another adaptivity recalc interval with even 0.01 difference, but 0.22 looks unusually large to me - curious what others say.



Good Luck

[maru]

Is till don't get one thing:

In the attached image, in the middle picture it says "noise level 2.5" yet the render stamp says "noise level: 2,49". Is this correct? So you set your noise limit to 2,5 but Corona stopped at 2,49?

If that's not the case, could you please make the tests a bit easier to understand:
1) Set noise limit to 2,5
2) Run rendering on PC1
3) Run rendering on PC2
4) Post the results showing number of passes and nose level reported by both computers in the render stamp?

Sorry if that's what you have already done, but the info is a bit confusing to me.

[Philip kelly]

So have a general question.

I generally sit and watch the render may be 75, 125, 200, 300 passes.
Should I be setting the noise limit to 2-2,5 and let that do the render time?

Thank you

phil

[maru]

(still waiting for more info from OP based on https://forum.corona-renderer.com/index.php?topic=18924.msg118994#msg118994)

So have a general question.
I generally sit and watch the render may be 75, 125, 200, 300 passes.
Should I be setting the noise limit to 2-2,5 and let that do the render time?

I am not sure if I understand the question right. You should use the limit of your choice. You are free to set all limits to 0, and then just decide when to stop rendering manually. You may use any of the limits to stop the rendering based on the specific value. You may also set up multiple limits and in such situation the rendering will be stopped whichever limit is reached first. So if you set up noise limit to 2 and time limit to 1 hour, the rendering will stop either after noise level reaches 2 or after 1 hour, depending on which if these happens first.

[Zray]

Is till don't get one thing:

In the attached image, in the middle picture it says "noise level 2.5" yet the render stamp says "noise level: 2,49". Is this correct? So you set your noise limit to 2,5 but Corona stopped at 2,49?

If that's not the case, could you please make the tests a bit easier to understand:
1) Set noise limit to 2,5
2) Run rendering on PC1
3) Run rendering on PC2
4) Post the results showing number of passes and nose level reported by both computers in the render stamp?

Sorry if that's what you have already done, but the info is a bit confusing to me.

yes, it's exactly what you mentioned. The middle image, I set noise = 2.5

[maru]

yes, it's exactly what you mentioned. The middle image, I set noise = 2.5

That's strange. Setting noise limit to 2,5 should stop rendering when it's 2,5 or more. Maybe there is some kind of "margin" for the noise limit, or it is rounded up in some way. I'll try to find out.
Sorry, that was a brainfart. Of course everything is fine. Rendering will stop if noise level is 2,5 or less, after the adaptivity recalculation is done.

But still, could you also do the test that I described in my previous message? (of course if you have time for this)
 

[Zray]

yes, it's exactly what you mentioned. The middle image, I set noise = 2.5

That's strange. Setting noise limit to 2,5 should stop rendering when it's 2,5 or more. Maybe there is some kind of "margin" for the noise limit, or it is rounded up in some way. I'll try to find out.
Sorry, that was a brainfart. Of course, everything is fine. Rendering will stop if noise level is 2,5 or less, after the adaptivity recalculation is done.
But still, could you also do the test that I described in my previous message? (of course if you have time for this)
 

I'm pretty busy this time. As I said, I've just been wondering about the different passes on 2 computers, it doesn't have any effect on my work so I hope this issue will be solved next hotfix.