Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Can nvidia-docker enable CUDA for images without specific drivers/support built-in? #766
Hopefully a quick question, I'm trying to get my head around nvidia-docker.. emby/embyserver is a container with a version of ffmpeg compiled for CUDA support.
I think from reading #531 that provided I use docker-nvidia wrapper (as opposed to just "docker"), and pass NVIDIA_DRIVER_CAPABILITIES=compute,utility,video as an environment variable, that nvidia-docker will mount the necessary libraries etc into my container "on-the-fly", and ffmpeg will "just work".
Sadly, this is not the case (I've probably misunderstood), and I get our old friend:
Further, libnvcuvid.so.1 is nowhere to be found on the container's filesystem.
Should I be able to do this, or should I be rebuilding an Emby container from nvidia/cuda as a base?
Worked a treat, thank you! :)…
-- Sent from my mobile device
On 16/06/2018, at 4:36 AM, Felix Abecassis ***@***.***> wrote: The variable that triggers our runtime or not is NVIDIA_VISIBLE_DEVICES=all, so you need to do something like that: $ docker run -ti --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=all -e NVIDIA_DRIVER_CAPABILITIES=compute,utility,video ubuntu:16.04 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.