Playblast with Gpu support not enough memory

[annovi]

I'm trying to use this image to run a Playblast render.
I’m running the container with GPU support (correctly enabled), but I cannot get it to work because it said there is not enough memory available.

I'm wondering if you experienced similar problem and you know how to solve it?

This is the log that I get:

File read in  0.32 seconds.
Viewport 2.0 floating point render target is turned off.
Viewport 2.0 floating point render target is turned on for color management.


——
VP2 Warning : Forcing GPU memory limit to 24576 MB. Results may differ from using detected memory limit.
VP2 Warning : Insufficient shader level capabilities. Vertex (3 vs 4), Geometry (0 vs 0) Pixel (3 vs 4)
VP2 Warning : Graphics hardware feature level insufficient to support renderer (2 vs 4)
Could not initialize Playblast renderer

Your help would be highly appreciated. Thank you in advance.

[mottosso]

Hi @annovi, thanks for posting this here. Playblasting in a Docker container has been the goal of many since I first made this container, and I sometimes hear rumours of someone succeeding, but still have not gotten to the bottom of it.

I remember in your email that you were running containers via NVIDIA's wrapper, which should forward the GPU into the container, at which point I would have expected it to work as-is, without modification. But, based on your struggles, it seems not.

• https://groups.google.com/g/python_inside_maya/c/sUZoqobbZTc/m/WKlm3lckCQAJ

This is the biggest thread on the topic, with some more clues here.

• https://groups.google.com/g/python_inside_maya/c/x9qZQiQLHkU/m/2vFEyJQRCQAJ

My understanding is that people ultimately resort to spinning up a non-VM machine, with a GPU, to perform their Maya playblasts on a headless machine. So if you, or anyone, does solve this I expect you'd be doing the industry a big favour!

Let's use this thread to collect all resources and experiences towards this goal.

[annovi]

Hi Marcus, Thank you for the extensive answer. For now we found a workaround, basically we render using the Hw acceleration and not using playlist viewport. Kind Regards, Andrea