OctaneRender™ Standalone 3.06 TEST

[coilbook]

Ok here are my initial tests, Adaptive Sampling does work just fine as you will see below,
All these are rendered with 7x1080's watercooled 2x Xeons e-5 2640 v4, at 1920x1080, pathtracing of course and the scene's only light source is a window with a light portal on the right and octane sun and sky env. Tems at max reached 44 C in a fairly hot room of 23 C with complete silence from the machine. also a shout to Smicha who build it ! I also did a super clean render at 8192 samples followed by the same quality with AS.

Here:

So based on your input the higher the Samples set the faster it cleans the noise? i need help in get this right somehow..do I need new drivers? on 375.63 now..

The explanation of AS is pretty simple.

First I describe how Octane works without AS:
Samples are shot from the camera through every pixel into the scene. Those samples are distributed uniformly over the field of view of the camera.

Advantages there:
The rendering is progressive. It clears up nicely and gives you really fast feedback.

Disadvantages:
If you have one spot in your picture that is really hard to render (due to high ray depth (glass, sss) or complicated sampling (chromatic aberration, indirect lighting) while every other part of the render has received enough samples you have to "over sample" those areas in order to reach the point where hard to render spot is noise free. So you loose a lot of render power to the spots where no further samples would be needed but have to receive them in order to overall reach the target sampling level to have a clean image even at the spot that was hard to clean up.

To solve this problem there is Adaptive Sampling:
With the "Noise Threshold" you set how much noise in the picture is OK. If the parts of the render have reached that noise level they will be cut off from active rendering. Leaving more power to the areas that have not reached that threshold.

Of course your rendering will go faster up to the specified sampling level because after some time there is less area to cover, fewer pixels to calculate. What it does not do at this point is render more noise free. It still only render up to the max specified SPP.

So the Workflow with AS is:
1: Make a render region that covers the area in your render where you have the most noise in your scene and see how many samples you need to clean it up.
2: Set this value as your max SPP level.
3: Turn on AS and set the noise threshold so that you are overall satisfied with the noise level of the green areas (that are excluded from active rendering)

So what happens is that the parts of the picture that do not need the samples get cut of at lower sample rates from the rendering. While the parts that need the samples will get up to the maximum SPP that you figured out before.
Since not the whole picture is covered with the high SSP number it normally renders quite quick.

It really depends on the scene though. If you do not have a big difference in noise in your scene (due to a lot of indirect light for example) AS can not do a lot. Because it can not turn off any, or only small areas from your render.

EDIT:
Just to clarify:
If you just turn on AS you will not get a more noise free image. What you get is a image that looks the same like the image with out AS but in a shorter amount of time. (depended on the scene)

I hope this was understandable.

Cheers,
Raphael

thanks. are there magic numbers for AS for animations. Doing this for every shot will take too much time

[Jolbertoquini]

Hi Guys,

Really great updates thank you for all hard work, I was testing the adaptive sampling is really nice. The only thing I think is is missing is a option to recognize the image luminance to help on the sampling, usually bright part and really dark areas the noise sometimes are not visible to much of corse only when we over expose on post. Maybe a extra option to help to clean more faster I gues still really good already with a small denoise we can't get quite clean renders.

Best,
JO

Just activate the Noise render layer, it will give information about the noise amount in light areas as well as in dark areas

Never mind I was not talking about noise preview dude...I did understand the explainations to be more clear and I didn't want give a name of render engine but I will now, I was talking a different way to analyse the noise such V-ray does also based on the image lookup but of corse as I messioned this may be a issue when extreme compositing are done....but never mind is already good already

[justix]

Oh wow guys! Thanks for Adaptive Sampling!!! Finally!!

I was bummed about having hardware issues and switching back to an older build machine with only 2 cards... but I have to say, good job Adaptive sampling may start normal but then starts speeding up like crazy!

I tried to mimic Vray's setting using 0.85 adaptive strength and does seem to be a sweet spot with 0.1 threshold, bumping threshold where needed. (previews -> finals) Good performance gains with the right settings. (obviously scene dependant)

Now that you have a noise render pass, it should be a step in the right direction for an internal Denoiser?

Benchmark - Idea PT Scene:

1500 x 1000
10 000 samples

No AS = 3min 43sec
AS = 1min 18sec

There is visually no difference and really clean. *win!*

.

Reading your settings I did test that scene too and with 10000 samples NO AS I get 23.10 min BUT with AS ON it went 9.52 min ( Gtx 780 ) really wish to be able to apply such cut to all my work... still I don't get any answers on why higherthe samples and more effective the AS seems to work or faster.. or iti s just my assumption based on my ignorance of this technique?

[Iceman9]

Reading your settings I did test that scene too and with 10000 samples NO AS I get 23.10 min BUT with AS ON it went 10.13 min ( Gtx 780 ) really wish to be able to apply such cut to all my work... still I don't get any answers on why higherthe samples and more effective the AS seems to work or faster.. or iti s just my assumption based on my ignorance of this technique?

With higher samples the AS threshold is attained in larger areas of your scene, thus you see more real-world benefit.

If you are using low samples AS never gets much chance to "kick in." Noise is still spread throughout the scene thus Octane still has a lot to calculate.

AS isn't a magic pill for all scenes and at all quality settings. It only works if/when it can reduce the amount of work the renderer must perform. Also AS doesn't do much at all in the initial phases of the rendering. It's benefit kicks in during the latter part of a render. With few samples Octane never gets deep enough into a render for it to give you much benefit.

[milanm]

@justix

The link that I shared explains AS in a way even a five year old could understand. Check it out.

Regards
Milan

[justix]

@justix

The link that I shared explains AS in a way even a five year old could understand. Check it out.

Regards
Milan

Yes thanks I had a good read although I'm tempted to re-calculate some of my recent work with As rising the samples..

[justix]

Reading your settings I did test that scene too and with 10000 samples NO AS I get 23.10 min BUT with AS ON it went 10.13 min ( Gtx 780 ) really wish to be able to apply such cut to all my work... still I don't get any answers on why higherthe samples and more effective the AS seems to work or faster.. or iti s just my assumption based on my ignorance of this technique?

With higher samples the AS threshold is attained in larger areas of your scene, thus you see more real-world benefit.

If you are using low samples AS never gets much chance to "kick in." Noise is still spread throughout the scene thus Octane still has a lot to calculate.

AS isn't a magic pill for all scenes and at all quality settings. It only works if/when it can reduce the amount of work the renderer must perform. Also AS doesn't do much at all in the initial phases of the rendering. It's benefit kicks in during the latter part of a render. With few samples Octane never gets deep enough into a render for it to give you much benefit.

That's what I thought..

[Zay]

After releasing the Region Render I get a lot of crashes when using Adaptive sampling.

More info on this.
When the slave is running the Master will crash a lot when releasing the region render by clicking in the viewport. Even if you shut down the slave and restart the Master, the Master is now very easy to crash using region render. A full reboot is required to get a stable Master and a working region render tool.

[haze]

Not sure if other met this issue but when adding an obj mesh the specular texture is not added anymore, RGB color is created instead.
Edit : And the IOR is set to 1 (if I'm right it was set to 1.3 before)

Could you send me an obj with this problem please?

[haze]

There is some nice stuff in this update. Thank you!

I was wondering if you would consider an additional "simplified" triplanar texture. Right now you need to set up 3/6 image nodes with the correct projection, then plug those into the triplanar node. If I have a folder full of textures I want to quickly try out on my mesh, I have to click the load button on 3/6 different rgb image nodes and browse for the image 3/6 times.

I basically need box mapping, with blending to remove the seams. Maybe it could be a new projection type for images too?

In v-ray the triplanar texture is much easier to use. I just plug 1 texture in, and the triplanar controls everything else (scale, blend, offset, rotation). If I want more control, it also allows separate textures as an option.

vray_modo_triplanar.PNG

Just so that I can make sure I understand correctly - would you mind sending me an orbx with the setup you are describing? Thanks :)