immich-podman-quadlets/docs/install/hwa.md

# HardWare Acceleration setup for immich with podman ## Nvidia If you are using NVIDIA on a redhat OS such as fedora or almalinux, kernel updates sometimes cause the nvidia-uvm and nvidia-uvm-tools to just... not load promptly... Use the following script and systemd file to correct this behaviour: You can ultimately place this wherever you want as long as it's referenced appropriately in the systemd file: nv_init.sh bash #!/bin/bash echo "modprobe nvidia" /sbin/modprobe nvidia echo "modprobe nvidia-uvm" /sbin/modprobe nvidia-uvm if [ "$?" -eq 0 ]; then # Count the number of NVIDIA controllers found. NVDEVS=`lspci | grep -i NVIDIA` N3D=`echo "$NVDEVS" | grep "3D controller" | wc -l` NVGA=`echo "$NVDEVS" | grep "VGA compatible controller" | wc -l` N=`expr $N3D + $NVGA - 1` echo "mknode nvidia" for i in `seq 0 $N`; do if [ ! -e /dev/nvidia$i ]; then mknod -m 666 /dev/nvidia$i c 195 $i fi done echo "mknode nvidiactl" if [ ! -e /dev/nvidiactl ]; then mknod -m 666 /dev/nvidiactl c 195 255 fi else echo "exit NVIDIA controller" exit 1 fi if [ "$?" -eq 0 ]; then # Find out the major device number used by the nvidia-uvm driver D=`grep nvidia-uvm /proc/devices | awk '{print $1}'` echo "mknode nvidia-uvm" if [ ! -e /dev/nvidia-uvm ]; then mknod -m 666 /dev/nvidia-uvm c $D 0 fi echo "mknode nvidia-uvm-tools" if [ ! -e /dev/nvidia-uvm-tools ]; then mknod -m 666 /dev/nvidia-uvm-tools c $D 0 fi else echo "exit nvidia-uvm" exit 1 fi echo "generate yaml in case of driver upgrade" nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml /etc/systemd/system/init_nvidia.service bash [Unit] Description=NVIDIA Init for CUDA After=network.target ConditionPathExists=|!/dev/nvidia ConditionPathExists=|!/dev/nvidiactl ConditionPathExists=|!/dev/nvidia-uvm ConditionPathExists=|!/dev/nvidia-uvm-tools [Service] Type=oneshot ExecStart=/bin/bash /opt/git/containers/rootless/containers/immich/nv_init.sh RemainAfterExit=yes [Install] WantedBy=default.target