I am using kdenlive-26.08.1-x86_64.AppImage on Debian 13 with a Nvidia RTX graphic card (Driver Version: 595.91.07 and CUDA Version: 13.2).
I often experience some instabilities (crashes, etc.) and most of the time just by performing some basic actions (reloading clips, moving clips on the timeline, etc.).
When looking into the logs (see an extract below) I can see many errors related to Cuda.
My questions :
1- Is Cuda natively supported by kdenlive?
2- If yes, are there any specific settings to respect (cuda version…)?
3- Concerning the errors bellow, any hint about how to fix them?
4- Workaround: is there a way to completely deactivate cuda and run kdenlive from cpu only?
Thanks a lot for your inputs!
Logs:
Generating audio levels for stream 0 of “/tmp/kdenlive-YOPOUr.mlt” using MLT
[h264 @ 0x7fdd394c5080] Hardware is lacking required capabilities
[h264 @ 0x7fdd394c5080] Failed setup for format cuda: hwaccel initialisation returned error.
MLT: “[producer_xml] failed to load producer \“0.502327:/media/…/PANA4392-25fps.mp4\””
[CUDA @ 0x7fd60ee6f140] cu->cuCtxCreate(&hwctx->cuda_ctx, desired_flags, hwctx->internal->cuda_device) failed → CUDA_ERROR_OPERATING_SYSTEM: OS call failed or operatio
n not supported on this OS
yeah kdenlive pulls cuda through mlt/ffmpeg when it sees a nvidia card. the appimage often fights the host driver though, which is what that CUDA_ERROR_OPERATING_SYSTEM + hwaccel init fail usually means.
quickest check: open Settings → Playback and turn off gpu / hardware acceleration, or launch with:
export MLT_AVFORMAT_HWACCEL=none
./kdenlive-*.AppImage
flatpak or the distro package usually behaves better with system nvidia than the appimage. paste nvidia-smi and say if you already tried flatpak.
I’m not sure this is true. I use Arch, and the repo-provided Kdenlive currently no longer allows me to use the GPU rendering feature. It used to work, now it doesn’t.
However, I recently tested the AppImage and GPU rendering is working again. Of course, this is a YMMV situation, but based on my experience with Krita as well, I’d say that KDE specialist/creative apps seems to work better as AppImages.
Though, of course, this is not the case for the OP. Maybe they need to try an older version of the Kdenlive AppImage.
fair, on arch the repo build can miss cuda too depending on how mlt/ffmpeg were built. the appimage still fights the host nvidia stack though, which is what that CUDA_ERROR_OPERATING_SYSTEM usually is.
for OP: still start by turning hwaccel off and paste nvidia-smi. if flatpak dies the same way, then its the driver/cuda side.
