Revision history for Rex-GPU
0.001 2026-03-29 04:20:30Z
- Initial release
- GPU hardware detection via PCI class codes (NVIDIA, AMD)
- NVIDIA driver installation (Debian 12/13, Ubuntu 22.04/24.04, RHEL/Rocky 8/9/10, openSUSE Leap 15.6/16.0)
- NVIDIA Container Toolkit installation with automatic repo setup
- CDI spec generation via nvidia-ctk cdi generate
- Containerd runtime configuration (RKE2, K3s, standalone)
- Support for Ubuntu 24.04: auto-detect nvidia-driver-NNN-server, handle virtual nvidia-smi package
- DPkg::Lock::Timeout=120 on all apt-get calls for resilience on fresh-boot systems
- All package installs use apt-get/dnf directly with dpkg/rpm verification
(Rex::Pkg dies on non-zero from DKMS/grub/initramfs post-install scripts)
- Reboot support with reconnect polling via Rex::LibSSH
- Tested on Hetzner dedicated servers with NVIDIA RTX 4000 SFF Ada Generation
(Debian 13, Rocky Linux 10.1, Ubuntu 24.04 LTS)
- eg/Rexfile example for GPU detection and cluster deployment
- Requires Rex::LibSSH for deployment to SFTP-less hosts