Closed findmyway closed 1 year ago
Will consider for the next update, thanks.
Will make adjustments together when the next version for selkies-gstreamer is out.
Note to self - additional backlogs for next update (MAKE SURE TO COMMIT ON EGL TOO):
The request has been done in NVIDIA/nvidia-docker#24. However, due to the following issues, the following section in the Dockerfile was required and may present issues the next time the base image is updated. As this is relevant to this issue, I am keeping this issue open. https://github.com/NVIDIA/nvidia-container-toolkit/issues/257 https://gitlab.com/nvidia/container-images/cuda/-/issues/158 https://forums.developer.nvidia.com/t/notice-cuda-linux-repository-key-rotation/212771/12
# Temporary fix for NVIDIA container repository
RUN apt-get clean && \
apt-key adv --fetch-keys "https://developer.download.nvidia.com/compute/cuda/repos/$(cat /etc/os-release | grep '^ID=' | awk -F'=' '{print $2}')$(cat /etc/os-release | grep '^VERSION_ID=' | awk -F'=' '{print $2}' | sed 's/[^0-9]*//g')/x86_64/3bf863cc.pub" && \
rm -rf /var/lib/apt/lists/*
Note that the MATE Desktop is incompatible with Xorg in the default repositories combined with NVIDIA drivers >= 495, therefore the desktop environment was changed to Xfce. Information: https://ubuntu-mate.community/t/can-anyone-here-upgrade-to-nvidia-driver-510-without-login-problems/25147/4 https://forums.developer.nvidia.com/t/cant-boot-after-upgrading-to-510-driver-on-ubuntu-20-04/202625/18 https://bugs.launchpad.net/ubuntu/+source/xorg-server/+bug/1959995
Is it possible for having GNOME desktop? As it’s much optimized than xfce and ideal for many Ubuntu operations?
Is it possible for having GNOME desktop? As it’s much optimized than xfce and ideal for many Ubuntu operations?
--init=systemd
, --cap-add=
, or is not based on Debian is unpractical to be used in Kubernetes).Things will improve when we start using Wayland, however.
A lot of these issues have been addressed in the latest commits. Closing. Ubuntu 22.04 will be supported after nvidia/cudagl image is released.
Current
cudagl
version is hardcoded in the dockerfile. How about turning it into anARG
and enabling it in the GitHub workflow?