OctaneRender™ 2020.1 XB1 [superseded by 2020.1 XB2]

[TRRazor]

There seems to be a problem with alpha-mapped objects casting shadows onto random walk sss-enabled geometry.
The shadow in this scene shouldn't be visible, since that part of the alpha map (read: the haircap) is transparent. Yet, it is casting a shadow.

Hi TRRazor,

Do you have a scene you can share for reproducing the problem?

Thanks.

Hi, wallace,

yes, sure.
Please check the scene attached. You'll find an orb surrounded by a torus. Although the torus' opacity is set to the value of "0" in the corresponding node, you will see, that there is still a visible faint line, being projected onto the orb, where the torus' geometry would be.
It's not precisely the problem, I mentioned in my original post (since there is no alpha map with areas with values of "0" in it) but it' s essentially the same problem.

[mojave]

Also for some scenes, when RTX is off, the scene renders fine. When RTX is on, the frame comes out as completly black. They're all exported from the C4D 2020 Plugin.

Thank you for sharing your scene, this was due to an issue happening when using AI light in combination with RTX which will be fixed in the next release.

[wallace]

I'm just doing some tests and noticed volumes are rendering wrong in 2020.

volume_wrong_in_2020.png

I have made a fix for this and it will be in XB2.

Hi, wallace,
yes, sure.
Please check the scene attached. You'll find an orb surrounded by a torus. Although the torus' opacity is set to the value of "0" in the corresponding node, you will see, that there is still a visible faint line, being projected onto the orb, where the torus' geometry would be.
It's not precisely the problem, I mentioned in my original post (since there is no alpha map with areas with values of "0" in it) but it' s essentially the same problem.

Hi,

Thanks for the scene.

I would like to ask the following questions after looking at the scene:
1. Are you also using high coherence ratio in your original scene?
2. What is the sample count you are using in that scene?
3. Does giving more sample with a lower coherence ratio allow the opacity 0 plane on the head converge to 100% transparent?

[mojave]

Very strange result on this scene.

scene_for_rtx.jpg

scene_for_rtx_2.jpg

2*2080Ti (NVLink)

Scene - > http://bit.ly/34CjKPn

Thank you for reporting this. We will look into it.

[TRRazor]

I'm just doing some tests and noticed volumes are rendering wrong in 2020.

volume_wrong_in_2020.png

I have made a fix for this and it will be in XB2.

Hi, wallace,
yes, sure.
Please check the scene attached. You'll find an orb surrounded by a torus. Although the torus' opacity is set to the value of "0" in the corresponding node, you will see, that there is still a visible faint line, being projected onto the orb, where the torus' geometry would be.
It's not precisely the problem, I mentioned in my original post (since there is no alpha map with areas with values of "0" in it) but it' s essentially the same problem.

Hi,

Thanks for the scene.

I would like to ask the following questions after looking at the scene:
1. Are you also using high coherence ratio in your original scene?
2. What is the sample count you are using in that scene?
3. Does giving more sample with a lower coherence ratio allow the opacity 0 plane on the head converge to 100% transparent?

Hi,

thanks for getting back with me so quickly.

That absolutely did it. I'm usually using coherence values around 0.35, since that was significantly speeding up render speeds before RTX acceleration came around.
I just set it to 0 and wasn't able to reproduce the problem. Same result with my original scene, the opacity now renders 100% transparent, instead of leaving those nasty shadows / lines.
Does that mean, however, that when using RTX, one can't use coherence?

I was using the following values in my scene (according to your questions):
1.) coherence value of 0.35
2.) 5000
3.) As mentioned above, it fixed it.

Thanks again!

[osiosiosi]

Question ...

Does NVlink work with 4 cards? Meaning a pool of 2 x 2 with total 22GB memory avaiable?

Can anybody confirm?

[gurel]

Is the standalone version the only one available or are the plugin modules linked anywhere?
I assumed the C4D plugin would be available since the press release makes a point of noise shader functionality.

Thanks

[SSmolak]

Question ...

Does NVlink work with 4 cards? Meaning a pool of 2 x 2 with total 22GB memory avaiable?

Can anybody confirm?

I can confirm that NVlink memory pooling doesn't work even for one pair at the moment. It probably works only for Quadro and Titan RTX. It doesn't work for any other Geforce card including 2080Ti.

If your scene extend memory of one card and you have disabled Out Of Core - Octane render will fail with out of memory error.

[wallace]

Hi,

thanks for getting back with me so quickly.

That absolutely did it. I'm usually using coherence values around 0.35, since that was significantly speeding up render speeds before RTX acceleration came around.
I just set it to 0 and wasn't able to reproduce the problem. Same result with my original scene, the opacity now renders 100% transparent, instead of leaving those nasty shadows / lines.
Does that mean, however, that when using RTX, one can't use coherence?

I was using the following values in my scene (according to your questions):
1.) coherence value of 0.35
2.) 5000
3.) As mentioned above, it fixed it.

Thanks again!

Hi,

It's not coherence ratio + RTX acceleration that is causing this issue, but the combination of coherence ratio + opacity is the main cause.

This is because the random numbers are more likely to fall within the same range, so the subsequent random numbers used for surface scattering and medium scattering are inherently different for surface area on the sphere that is being occluded by the transparent torus vs the non occluded area.

This just means that you will need more samples for that difference to go away, generally the guide line is that the higher coherent ratio you use, the less variance you will have on your image, but the required sample count is also higher for you to converge to the correct image.

Thanks,
Wallace

[karl]

Is the standalone version the only one available or are the plugin modules linked anywhere?
I assumed the C4D plugin would be available since the press release makes a point of noise shader functionality.

Thanks

Please see this thread for the C4D plugin: viewtopic.php?f=85&t=73352