OctaneRender™ 2020.1 RC4 [superseded by 2020.1] Tweet

[mojave]

one of our slaves just crashes when using RC versions. Slave has 2080tis ans titans mixed with cuda error and failing kernels. The slave that has all Rtx cards 2080ti works fine

What version do you use and could you please share a scene that reproduces this issue for you? We have addressed an issue causing an error like this in RC4 so it is worth trying if you haven't done so already.

HI
We were using RC2 since max doesnt have RC4. We also get optixpipline error. I will try to send you the scene tomorrow as PM

Thanks, please let us know if this still happens with RC4 as it looks very much as one of the issues that already got fixed.

[ptunstall]

We too are getting this memory error with a scene that was previously working in RC1(was disabled today/this weekend). I can't share the ORBX file publicly but I have a dropbox link if I could PM/email it.

GPUs that it fails on 1080/1080ti, quadro RTX 4000, titan RTX, and titan V.

We have RTX OFF.

Ours fails on both centos and windows.

We rolled back to RC2 and this appears to be working correctly for now. Nope RC2 is failing as well... Is there a way to get RC1 back online?

After thinking, this might be a volume issue. 2 weeks ago we had to do some hackery with converting an emissive(fire) VDB to a point cloud to get it to properly light a scene. It'd fail in a similar fashion or simply just stop rendering mid way through. I saw you guys had fixed similar issues in the bug fix notes so we assumed it had been fixed. Up to this point I've been using randomwalk for ALL my volumes since then except for TODAY where I switched to a standard medium. Happy to share ORBx and VDB files just need to send privately.

Could you please confirm this happens also with RC4 and share any ORBX we can use for testing if that's the case? Thank you.

Yes, sorry, this is happening in RC4. Can I PM you dropbox links to the ORBx files? I cannot share this work publicly.

[mojave]

We too are getting this memory error with a scene that was previously working in RC1(was disabled today/this weekend). I can't share the ORBX file publicly but I have a dropbox link if I could PM/email it.

GPUs that it fails on 1080/1080ti, quadro RTX 4000, titan RTX, and titan V.

We have RTX OFF.

Ours fails on both centos and windows.

We rolled back to RC2 and this appears to be working correctly for now. Nope RC2 is failing as well... Is there a way to get RC1 back online?

After thinking, this might be a volume issue. 2 weeks ago we had to do some hackery with converting an emissive(fire) VDB to a point cloud to get it to properly light a scene. It'd fail in a similar fashion or simply just stop rendering mid way through. I saw you guys had fixed similar issues in the bug fix notes so we assumed it had been fixed. Up to this point I've been using randomwalk for ALL my volumes since then except for TODAY where I switched to a standard medium. Happy to share ORBx and VDB files just need to send privately.

Could you please confirm this happens also with RC4 and share any ORBX we can use for testing if that's the case? Thank you.

Yes, sorry, this is happening in RC4. Can I PM you dropbox links to the ORBx files? I cannot share this work publicly.

Yes please, thank you for your help.

[coilbook]

Can anyone from otoy explain about rtx?
we have 25 cards (20 are 2080TIs and 5 Titans)
RTX OFF: 37 sec
RTX on: only 35 sec
The scene is 5 mil polys.

Do RT cores only care about the geometry complexity no matter how many lights are in the scene?

Also AI light.
AI light on or off does nothing. Same noise same render time even with hundreds of light sources. Sometimes AI slows down the system but no visible results

Thanks

[mojave]

Can anyone at otoy explain about rtx?
we have 25 cards (20 are 2080TIs and 5 Titans)
RTX OFF: 37 sec
RTX on: only 35 sec
The scene is 5 mil polys.

Do RT cores only care about the geometry complexity no matter how many lights are in the scene?

Also AI light.
AI light on or off does nothing. Same noise same render time even with hundreds of light sources. Sometimes AI slows down the system but no visible results

Thanks

It would depend on your specific scene, for instance non-triangle primitives, texture displacement as well as volumetric effects will not benefit from RT cores as the HW is designed to deal with regular triangle data. In some scenarios material information such as dirt or accurate rounded edges currently require to fallback to CUDA (this is an improvement we plan for 2020.2).

If you could please share a repro scene for this as well as for the CUDA error you reported via PM we could probably provide more information.

[wallace]

Also AI light. AI light on or off does nothing. Same noise same render time even with hundreds of light sources. Sometimes AI slows down the system but no visible results

This is scene dependent. Noise in general can be contributed from direct bounces and indirect bounces. If the indirect bounces' contribution is high then AI light does not help much as AI light only addresses issues in direct light sampling.

[kmdani]

I'm not sure, if it is a render bug, or what, but I'm getting bad normals on rendering a sequnce. (My geometry is different on each frame, it is "growing).
Here is a good frame:
https://drive.google.com/file/d/1F1zAu-d40fuyHRo_O-hMmidm8dFknsKp/view?usp=sharing
Here is a bad frame: black artifacts appear on the geo.
https://drive.google.com/file/d/1E6_SKaS7T6HXe3Y_AFF6aLBq5T4qhjkI/view?usp=sharing


I thought that there is something wrong with the geo, but normals look fine, also, made sure to get a v=0 (not to mess up the motion blur). But I realised that if I set Octane from Scene update to Full scene reload, it fixes it.
Also, while having the Scene update option, the artifacts were pretty random. After rerendering a frame, they went away.

I know it is beacuse of the different point count per frame, but any solutions? Until then, I guess just leave it on Full scene reload.

[Notiusweb]

AI Lighting is non-functional with Path Tracing kernel?

[Notiusweb]

Wallace - can I stack layers indefinitely with Composite???!!!!!
Please say YES!!!!!

Yes.

Here're the differences between layered material and composite material.

1. Composite material works like traditional mixed material, where you stochastically sample one of the layers and evaluate it.
2. The above means that you only evaluate one layer at each bounce intersecting a surface, so it's much more lightweight. The layered material evaluates the entire stack all at once to calculate the correct attenuation between layers and also combines the contribution of different BRDFs. This makes it less noisy.
3. Composite material weights the layer using a user-specified texture, you can use things like fresnel node to weight a specular material on top of a diffuse material, but this is less accurate than layered material. This is because layered material uses actual fresnel (i.e. microfacet fresnel with half vector) for actual weighting and attenuation.

So if you do things like 10 diffuse decals, then you don't really care about the accuracy of weighting of layers and attenuation, and in that case you should use composite material.

Thanks for taking the time to discuss this. The composite is offering me much more than I realized!!!!

[linograndiotoy]

It looks like Denoiser is not working properly on the background in RC4 if a Volume Medium is used.

Beauty:

Denoised:

Denoiser_Background_issue.ocs

(75.52 KiB) Downloaded 202 times