Raise sample or resolution for better renders Tweet

[haffy]

Hi,

I am testing a bit here, I have a scene with sheats with small details in them. When I zoom in I get a perfect result, but when I zoom out like 2m I lose it all.
What is best for getting noise-free detailed images?

I have rendered the scene out with this setting:

3840x2560 px
Max samples: 50000
Diffuse depth 4
Specular depth 4
Scatter depth 2

Took about 15 hours.



And this render is rendered with these settings.

1920x1080 px (this is the target resolution)
Max samples 4096
Diffuse depth 12
Specular depth 12
Scatter depth 8




Test scene with shaderball.

This one is:

Max samples: 2048
Diffuse depth 8
Specular depth 8
Scatter depth 4



This one is:

Max samples: 2048
Diffuse depth 8
Specular depth 512
Scatter depth 4



Or is thsis "adaptive noise" green-thingie not somthing to use for see how noise free the render is?


Another test, cranked up the settings.




Questions.

How much are too many samples in both Max Samples and Diffuse/Specular/Scatter?

And shall I raise the samples in the lights too, to get better noise-free-renders?

Does Octane remove details when farther away in some way?

[elsksa]

Hi haffy,

How much are too many samples in both Max Samples and Diffuse/Specular/Scatter?

As it is scene dependent, it is impossible to indicate what amount of samples is needed. However, in an optimized scene and depending on the hardware, the average is ranging from 512 to 4096.
Warning: these are not samples but ray depth parameters, respectively for diffuse (self explanatory), specular (reflection, refraction) and scatter (e.g. volume, SSS).

And shall I raise the samples in the lights too, to get better noise-free-renders?

Warning: the "Light Sampling Rate" is not a light sample option. It is a "light sampling" priority. As an example, a scene with two area lights, if light A has a Light Sample Rate of 1 and Light B of 2, the light B sampling will be prioritized.
It is mentioned here, which I highly recommend to consult:https://www.elsksa.me/scientia/cgi-offline-rendering/octane-render-settings

Does Octane remove details when farther away in some way?

As in reality with a digital camera, for the same resolution, the further away, the less the details will be maintained. In a digital camera, other elements can affect it, such as the OLPF (Optical Low Pass Filter) and raw debayering algorithms (to just name a couple) are also taken into consideration, but that is off topic.

In rendering, there are a few ways to approach it:
a higher resolution,
lower filter value in the render settings and perhaps,
you could consider disabling texture-filtering: https://render.otoy.com/forum/viewtopic.php?f=30&t=68094#p341556.

Or is thsis "adaptive noise" green-thingie not somthing to use for see how noise free the render is?

Not exactly. Here is an explanation of how does it work: https://youtu.be/dHTnaYjj3g4

Side note, it is strongly recommended to use an appropriate display rendering transform. More information: https://www.elsksa.me/scientia/cgi-offline-rendering/octane-render-color-management and
https://www.elsksa.me/scientia/cgi-offline-rendering/cgi-color-management-survival-kit

[haffy]

Thanks Elsksa, will sit down and read those links tonight. But I could not find anything regarding Diffuse/Specular/Scatter depth. Is that now samples, but ray depth? But should I not se much more difference between 8 and 512?

In Clarisse samples were multiplied on the value in Diffuse/Specular/Scatter, is that not the same here?

[elsksa]

I knew this username was familiar. I must have seen you on the Clarisse community Discord server or forum.

Octane is quite differently simpler.
Octane does not have per material or per light sample count, neither in the kernel settings.
You could think of it as AAA in Clarisse (prior to Angie, at least), since when "enabled", per material and light sample counts are becoming irrelevant.

Regarding the ray depth, it is mentioned in one of the links I've shared.
https://www.elsksa.me/scientia/cgi-offline-rendering/octane-render-settings
You've probably missed it.

The equivalent in Clarisse is:

clarisse_zkAcEEWR0o.jpg (77.41 KiB) Viewed 1069 times

It says "Sampling > Path Depth" but it is not directly about sampling and instead, defines the maximum depth bounces of a given path (diffuse, reflection/refraction, volume/SSS, etc). For instance, a value of zero for diffuse is equal to disabling diffuse illumination. Therefore, increasing it will add more "bounced light" to the scene.

These values should be at minimum 2 or 3 for plausible results. Some scenes may require the value to be higher e.g. 10 or more. It is also scene dependent.

[haffy]

Yes, that is true. I was active there a couple of years ago. Very nice community and software, but too big/clunky for me as a freelancer/single-artist at this stage.

Yes, I missed it. I was in a hurry to send a scene to RNDR and just did a page search and missed the - in-between ray and depth.

So Max Samples is like AAA in Clarisse, that's nice. The more the better (or brute force) then.

Ray-depth.
If I fast render (without the lock) and raise with +1 each time, I will quickly see when enough is enough. Or do I need to render like 5000 samples at high res to see it? Yes, I can and will use the render-region tool.
That sounds about right, if so.

So in hard terms a simple scene lighted with like HDRI and backplate could only need like max.samp 2048, diffuse 4, specular 4, sss 0 (if there is none) and an indoor scene lighted through a window could need max.samp 15000, diffuse 12, specular 12, sss 0 (if there is none). About right?

Regarding your very nice manual, guide. The "GI Clamping" section seems to be missing some image or video, showing the differenses. Would love to see this, as I think I may be using this wrong too.

[elsksa]

Ray-depth.
If I fast render (without the lock) and raise with +1 each time, I will quickly see when enough is enough. Or do I need to render like 5000 samples at high res to see it? Yes, I can and will use the render-region tool.
That sounds about right, if so.

The resolution is not related to the ray depth. To find out the appropriate ray depth value with this simple technique, feel free to use any resolution.

So in hard terms a simple scene lighted with like HDRI and backplate could only need like max.samp 2048, diffuse 4, specular 4, sss 0 (if there is none) and an indoor scene lighted through a window could need max.samp 15000, diffuse 12, specular 12, sss 0 (if there is none). About right?

As mentioned, it is not easily predictable. Each scene is different in their own ways. There are some rule of thumbs, though. Such as in Automotive, a detailed headlight will often require high specular ray depth, or interior higher diffuse depth.

Regarding your very nice manual, guide. The "GI Clamping" section seems to be missing some image or video, showing the differenses. Would love to see this, as I think I may be using this wrong too.

I've been told about this issue yesterday!
It seems to be a bug/issue only on chromium based web browsers (Chrome, Brave, etc). It does work as expected in, for instance, Mozilla Firefox: