davincibox: CUDA stops working permanently after container restart

IMPORTANT

Before reporting a bug, make sure that you can reproduce it on the latest version of davincibox.

See the upgrading section of the README for more information.

Also, be sure to read the Disclaimers section of the README first. Your issue may already be known and/or is an upstream issue with DaVinci Resolve directly, not with this project.

Describe the bug

A clear and concise description of what the bug is On latest Bazzite, with an NVIDIA GPU, after installing Resolve with the provided ujust command, CUDA integration is fully functional on first launch, but completely stops working any launch after that. This has been an issue for more than 2 weeks for some users

Installation method used

  • setup.sh
  • Manual
    • [+] Distrobox
    • Toolbox

To Reproduce

Steps to reproduce the behavior:

  1. Go to ‘Memory and GPU’ on initial box launch
  2. See CUDA selected in GPU processing mode
  3. Restart box
  4. See CUDA gone, only OpenCL is selectable

Expected behavior CUDA should fully work

A clear and concise description of what you expected to happen.

Screenshots If applicable, add screenshots to help explain your problem.

Image

Image

System information (please complete the following)

  • Distro: [Linux 6.13.5-102.bazzite.fc41.x86_64]
  • Desktop Environment: [KDE]
  • GPU: [NVIDIA GTX 1660 SUPER]
  • Davincibox version: [ghcr.io/zelikos/davincibox latest 144ebfbfee04 16 hours ago]

Additional context Add any other context about the problem here.

About this issue

  • Original URL
  • State: open
  • Created 2 years ago
  • Comments: 18 (7 by maintainers)

Most upvoted comments

Update: #167 does not fix this or other linked issues. However, this issue in the nvidia-container-toolkit repo sounds similar to what’s happening here: https://github.com/NVIDIA/nvidia-container-toolkit/issues/48

If that issue is the same as here, then we are still waiting for an NVIDIA driver update as I originally thought

Neither. It’s likely an NVIDIA driver bug

tested davincibox on davinci resolve studio 20 with rpmfusion and akmod-nvidia installed and it seems to be up and running again 🤔

I tried this too, davinci opens up and even recognizes the cuda cores and is also ready to use. But sadly when using it it is way choppier. I think this is because for some reason the gpu is not used properly.

I played a clip in a newly installed container with the --nvidia flag and then did the same after stopping and restarting the container and installing akmod-nvidia.

This is my cpu and gpu usage (and memory) on the fresh install. Playback runs smoothly and the gpu is used a lot while the cpu is used only a bit:

Image

And this is my cpu and gpu usage after restarting the container and installing akmod-nvidia. This time playback is very choppy and the gpu remains almost untouched while the cpu is doing all the work:

Image

I also tried every combination i can think of: Using the nvidia-container-toolkit instead of the --nvidia flag, using both, installing akmod-nvidia on top of it etc. but the result was always choppy playback and a maxed out cpu or it didn’t work after restarting the container.

Other things i found weird but since i’m no computer expert i don’t really know what to make of it:

  • After restarting a container with the --nvidia flag running nvidia-smi looked the same as on the first start.
  • installing davinci after restarting the container resulted in a non-working davinci directly.

tested davincibox on davinci resolve studio 20 with rpmfusion and akmod-nvidia installed and it seems to be up and running again 🤔

Image

A potential fix is being implemented in #167. To test, follow the manual setup steps for davincibox via distrobox, but replace --nvidia with --additional-flags "--gpus all", i.e.:

distrobox create -i ghcr.io/zelikos/davincibox:latest --additional-flags "--gpus all" -n davincibox