Need some advice from more experienced folks about settings and render times

[Correntes]

Hello,

So this is my first test. Its directly converted from vray to corona.
My objective is to compare render times to see if I can get corona in production pipeline here at the office.

I used PT - HD, standard settings and let it cook for about 3h:13min in a dual xeon. This is what I got... plenty of noise.
3 hours of processing aren't enough to clean a scene of this type ?

Mind you that this is an optimal interior (the back and side of the room are windows) and its lit only with a HDRI.

Hope I can get some insights about it.


Thanks

[rafpug]

Hello Correntes

Can you describe the combination of lights that you used for the scene?

Thanks
Raf

[Correntes]

Hello rafpug

Sure...

I only used hdri on max environment slot.

Thanks

[Correntes]

So I did a region test to see if how many passes are needed to clean that wall.

I let it run for about +- 900 passes to get acceptable noise levels. Well that would mean about +- 18h render...

[Ondra]

something is definitely wrong. Are you using default settings (PT samples, lights multiplier)? Can you printscreen the stats window (you activate in in VFB) after some time (~10 passes) when rendering?

[Correntes]

Hello Keymaster,

I tried corona in another project and had similar behaviour,  I tough that was due the fact that the other project have smaller windows.


Here it is a screen-shot of the settings from this current project.

Settings for what I know, are the default ones.



thanks

[Juraj]

What HDRi do you use ? High in contrast ? Big resolution ?

[Correntes]

Hello Juraj,

Its one HDRI from Peter Guthrie.

I'm using it with gamma boosted to 0.75. And I'm running it from a colorcorrection map and vrayhdri map.
Due this setup never gave me problems in the past I never thought that could be cause.

I'll guess that I can start by trying using a cleaner solution.


I will report tomorrow with my findings.


Mean while if you find something awkward with the settings please alert me.


Thanks

[Correntes]

Hello,

I tried to swap hdri for coronasky lightning went worse as expected... similar behaviour maybe a little less grainy.

Then I turned off the glass (a box with a material applied with refraction onesided (box)) i got acceptable noise levels by the 327 pass... That would mean more or less a render of 6 hours


Better but I think that for this kind of scene the render times are still to high and with allot of conditions (glass turned off, coronasky)

[Ludvik Koutny]

Hello,

I tried to swap hdri for coronasky lightning went worse as expected... similar behaviour maybe a little less grainy.

Then I turned off the glass (a box with a material applied with refraction onesided (box)) i got acceptable noise levels by the 327 pass... That would mean more or less a render of 6 hours


Better but I think that for this kind of scene the render times are still to high and with allot of conditions (glass turned off, coronasky)

There is no reason to do any compromises when it comes to the way scene looks. So i definitely would not replace HDRI with corona daylight. Does not make any sense. Changing the look of your scene completely is not a proper way of optimizing scene performance.

Problem in your scene is probably glass. And it has been mentioned at least 100 times already. For glass in windows, use either hybrid refraction mode, if you glass geometry has any thickness. Or twosided mode, if your glass are just single planes. If you leave refraction mode set to solid, then anything behind the glass (inside of the room) is computed as a caustics, not direct lighting.

[Polymax]

Can you post your scene? I will test it!

[Correntes]

Ahrg... I already had saw the" hybrid refraction mode" thing here in some threats and I completely mistaken for onesided in this scene  =|

[Ludvik Koutny]

Ahrg... I already had saw the" hybrid refraction mode" thing here in some threats and I completely mistaken for onesided in this scene  =|

Hybrid and twosided modes are fast ones. Solid is one with caustics. If you use hybrid or twosided and still have slow rendertimes, then there will be some other problem in your scene :)

[Correntes]

Can you post your scene? I will test it!

Hello Polymax, thank you but unfortunately I can't post the scene.


The glass in hybrid certainly helped thanks for the reminder Rawalanche .

3 hours render, PT-HD standard settings, noise is somewhat acceptable, not perfect but acceptable. Can it be optimized ?


ANother question:

Corona framebuffer is 32 bits right ? I saved the image in exr and I felt that in a way I didn't got that dynamic range that I usually do.

It could be for the position of hdri but when I played with curves I couldn't boost mids like usually.
I didn't compressed highlights in corona post processing  to keep max color info but I used exposure compensation to correctly expose the scene.

Thanks

[cecofuli]

For 3500px render  (A4 300 DPI) usually I render in  7-10 hours...   (with I7 970  @ 4 Ghz)

[Ondra]

did you do any contrast enhancement? Or better, can you printscreen the post processing tab?

3 hours seems... working. Certainly not the best result, but it is in the slower end of usual render speeds, given your CPU. Yet you have very low rays/s. There could be several reasons, for example look here: http://forum.corona-renderer.com/index.php/topic,1662.0.html
Are you using any portals?

[Correntes]

Cecofuli:

Yeah :S

Well it's getting rare the works that have manageable deadlines. Usually we have 1 day for 1 image. So if we have to do 4 images, we have 4 days.

That's why I'm trying to migrate to corona. Less settings to mess about, less setting tests to optimize scenes, more time to work on what it real matters.


Keymaster:

Not using portal lights .

Here it is the post-processing settings screen.

I had to use -4 exposure compensation to correctly expose the image. Should I use it only to preview and when saving the image revert back to 0 ?

Thanks

[rafpug]

Hello Correntes

The value of Lights samples Multiplier what corresponds ?

[Correntes]

Hello Correntes

The value of Lights samples Multiplier what corresponds ?

hello,

Everything default.

Screenshot in Reply #5


Thanks

[rafpug]

this value corresponds to 2?

[Correntes]

Yes

[rafpug]

try to render on a part where there is more noise "Crope" bringing the value to 20/25...in extreme case even 40

[rafpug]

try to render on a part where there is more noise "Crope" bringing the value to 20/25...in extreme case even 40

sorry,

Crop -  preview render

[Ondra]

no, this settings is fine, you are not losing any dynamic range because of that

[rafpug]

Hi Ondra,

which dinamice range? in what sense?

[Ondra]

I mean tones beyond displayed white. You partially lose them every time you set highlight compression over 1 or enable contrast

[rafpug]

I understand !

[rafpug]

I mean tones beyond displayed white. You partially lose them every time you set highlight compression over 1 or enable contrast

...should always have White balance as a value of  |k| 6500

[rafpug]

Ondra, a curiosity

When you switch from a Secondary solver to another,
and change the settings

- for example by  PT+PT  to  PT+VPLS

VPLS and change the value Emitted count to 1000

returning with the PT + PT method, the processing of rendering recognizes the 1000 VPLS?

[Correntes]

Yeah... Confirmed that I get the same results if I use -4 exposure on post-processing corona tab or after effects exposure.

rafpug:

With value 20 in  Lights samples Multiplier I get a clean noise free crop. 160 passes 5min render it doubled the render times.

With value 2 160 passes 2.5 min I got noise.

So I tried to run the crop with value 20 for 2.5 min. I got more or less 70 passes with a little less noise than with value 2.


My conclusions are that upping that value, I get slower passes but with more quality.
If I let the render cook for 3 hours I will get less passes but based on the crop test it will have a little less noise.


EDIT

I used a value 40 in Lights samples Multiplier. I let the crop run for 1.5 min (33 passes) and I got more or less the same results that 20 light samples multiplier cooking for 2.5 min.


Is there any drawback in use always 40 in lights samples multiplier ?

[rafpug]

Yes Corrents

I wanted to confirm if the noise was gone!

Rendering time you can also halve if you adjust the values ​​of the secondary solver depending on your scene

You need to do 3 4 tests on a sample with the Crop and calculates the time you!

Greats
Raf

[Correntes]

Thanks for the input rafpug, can you explain a little your process to optimize a scene ?

[rafpug]

EDIT

I used a value 40 in Lights samples Multiplier. I let the crop run for 1.5 min (33 passes) and I got more or less the same results that 20 light samples multiplier cooking for 2.5 min.


Is there any drawback in use always 40 in lights samples multiplier ?

best not to overdo it too  ^ _ ^

[rafpug]

Thanks for the input rafpug, can you explain a little your process to optimize a scene ?

sure Correntes
also a board uses the method PT+HD which is much faster both in the rendering of the image in the rendering time

The setting of secondary solver depends from one scene to the other one and go calibrated.

In yellow the values ​​that you should not change

The Next
Raf

[Correntes]

Thanks rafpug


What about the ones I can change, what they do ? And what the guidelines to change them ?

[rafpug]

Hello Correntes

1 Steps -You can do a test checking the time of the first passes?

...Time Limit

[rafpug]

Thanks rafpug


What about the ones I can change, what they do ? And what the guidelines to change them ?

For explanations about the settings of the values, you have to ask the staff because they are very professional.

more info here:
http://corona-renderer.com/doc/