OctaneRender™ 2020.1 RC1 [superseded by 2020.1 RC2]

[mojave]

Thanks for new release guys, good work on many fixes.

But from what i see my "sphere.orbx" report in 2020.XB3 is still broken in this 2020.1 RC1 although it's been reported as fixed.

Check these screengrabs, LEFT is correct 2019.1.5 and RIGHT is 2020.1 RC1 render = BAD.
In XB3 it was completely black so this is and upgrade/better but still not correct.

Please fix that.

Thanks

Thank you for the heads up. Your material uses vertex displacement, which is not applied in either version as the mesh has no UV map but they should not look different so we will look into that.

[mojave]

NVLink memory pooling for RTX 2080ti still doesn't work

We will look into this, thank you.

[mojave]

the 2080 ti just report error please see below images

Can you share what driver version you are using please?

[jimho]

the 2080 ti just report error please see below images

Can you share what driver version you are using please?

Hi Mojave,
Thanks for the prompt respond.
I am using the newest Nvidia studio driver on that slave, it is 442.19, it was just installed before the the new RC1 release,

Jim

[dynaraton]

Memory Pooling seems to be working on my twin RTX Titans.

Question - shouldn't my P2P box get full before I crash?
I crash if I add anymore geometry (please see picture below)
Over 138 Million Triangles (8k x 8k output)

Also, am I getting the combined 9200MB + 9036MB for a total of 18.2 GB of memory pooled Vram? Maybe the P2P box not being full is confusing me.

Also, I had to turn up my GPU headroom to max 24,576 MB to get this to render.
Out-of-Core off
RTX acceleration, De-noise, Post effects, Displacement, Emitters etc. All ON and they seem to work like a charm.

Thanks in Advance.

[karl]

Memory Pooling seems to be working on my twin RTX Titans.

Question - shouldn't my P2P box get full before I crash?
I crash if I add anymore geometry (please see picture below)
Over 138 Million Triangles

Also, am I getting the combined 9200MB + 9036MB for a total of 18.2 GB of memory pooled Vram? Maybe the P2P box not being full is confusing me.

Also, I had to turn up my GPU headroom to max 24,576 MB to get this to render.
Out-of-Core off
RTX acceleration, De-noise, Post effects, Displacement, Emitters etc. All ON and they seem to work like a charm.

Thanks in Advance.

Currently, only some aspects of the scene (geometry data) are able to use pooled memory. So you may still have a lot of data in your scene (e.g. textures) that needs to be resident on every device (or out-of-core) which would mean you may run out of memory despite using pooled memory for as much data as possible. We are keen to support memory pooling for more types of data in the future.

The numbers you are seeing mean that the first device is using 9200 MB of data stored on the second device, and the second device is using 9036 MB of data stored on the first device. It's not so much a shared "pool" as it is that one device can make use of data stored on another device.

I suspect the reason why you needed to increase the GPU headroom so much is because of RTX acceleration being enabled. RTX can use a lot of memory, and using it in combination with NVLink may cause problems for some scenes at the moment. We are looking to improve this in future releases.

[dynaraton]

Memory Pooling seems to be working on my twin RTX Titans.

Currently, only some aspects of the scene (geometry data) are able to use pooled memory. So you may still have a lot of data in your scene (e.g. textures) that needs to be resident on every device (or out-of-core) which would mean you may run out of memory despite using pooled memory for as much data as possible. We are keen to support memory pooling for more types of data in the future.

The numbers you are seeing mean that the first device is using 9200 MB of data stored on the second device, and the second device is using 9036 MB of data stored on the first device. It's not so much a shared "pool" as it is that one device can make use of data stored on another device.

I suspect the reason why you needed to increase the GPU headroom so much is because of RTX acceleration being enabled. RTX can use a lot of memory, and using it in combination with NVLink may cause problems for some scenes at the moment. We are looking to improve this in future releases.

Thank you Karu for the explanation. It is a great start with memory pooling and looking forward to future improvements. Also, I do not see any noticeable slowdowns with memory pooling...which is great also.

[grimm]

RT cores provide acceleration for just triangles at this stag. I think you wanted to share an ORBX package instead?

Thanks, that makes sense. Did the OCS file not work? I'm still curious as to why the file doesn't render as fast as the 4.05 version of SA does? That way I know how to adjust it for the new version.

Jason

[enricocerica]

Hi,
With complex scenes, I often manage to create groups limited to the visible objects for the view angle to reduce the amount of used memory. But this could quickly becomes a real mess :-/
What would be really useful in such situation is to have a render target option to be able to exclude meshes in the group linked to that render target. Working this way, I could link the same group to all my render targets and just select which mesh (object) I want to use for each angle view. It would really simplifies the workflow !

[mojave]

RT cores provide acceleration for just triangles at this stag. I think you wanted to share an ORBX package instead?

Thanks, that makes sense. Did the OCS file not work? I'm still curious as to why the file doesn't render as fast as the 4.05 version of SA does? That way I know how to adjust it for the new version.

Jason

The OCS itself did work but it doesn't contain any assets This is why I asked for the ORBX for testing what you report. Thanks for your help.