Right, nvidia-container-cli is specifically designed to use files from the host (outside of snap environment) as the files it loads (through dlopen) cannot be bundled (cuda, driver files, ...).
nvidia-container-cli has logic to effectively chroot prior to processing any of the dlopen. It's then expected that the driver libraries on the host are generally built conservatively and can be loaded even by a slightly older version. It's true that moving LXD to core22 would certainly solve the error here, though it would trade it for another problem, which is that those same libraries, which nvidia-container-cli passes through to the container would then only work on 22.04 containers or up. On amd64 22.04, this all works, including passing through the driver libraries and binaries from the host to a container as old as Ubuntu 18.04. So something weird happened with the equivalent arm64 build here. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1971978 Title: Driver binaries fail to load on arm64 through LXD To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-510/+bug/1971978/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs