NVIDIA-MIG-MANAGER(1)

💅 NAME 💅

nvidia-mig-manager — bestie, it's quite literally nvidia-mig-manager. That's the whole name. Iconic of it tbh.

---

✨ DESCRIPTION ✨

okay so. sit down, grab your iced coffee, we're doing the nvidia-mig-manager rundown 🫶

🎀 NAME:

nvidia-mig-manager — described in the original as "A new cli application". a brand new lil cli app. freshly born. main character energy fr.

🦋 USAGE:

nvidia-mig-manager [global options] command [command options]

that's the shape of it: global options first, then a command, then that command's own options. no cap.

⭐ VERSION:

unknown

yeah it literally says unknown. we don't know her version and she's not telling. 💀

👑 COMMANDS:

• help, h — shows you a list of commands, or help for one specific command. the little tour guide of the operation.

🌸 GLOBAL OPTIONS:

• --kubeconfig value — absolute path to the kubeconfig file. env var: $KUBECONFIG. absolute. like, full path, bestie, don't be cute with a relative one.
• --node-name value, -n value — the name of the node to watch for label changes on. env var: $NODE_NAME. she's watching that node's labels like they're a storytime.
• --config-file value, -f value — the path to the MIG parted configuration file. env var: $CONFIG_FILE.
• --reconfigure-script value, -s value — script to run to do the actual MIG reconfiguration. default: "/usr/bin/reconfigure-mig.sh". env var: $RECONFIGURE_SCRIPT.
• --host-root-mount value, -m value — container path where the host root directory is mounted. default: "/host". env var: $HOST_ROOT_MOUNT.
• --host-nvidia-dir value, -i value — host path of the directory where the NVIDIA managed software directory is typically located. default: "/usr/local/nvidia". env var: $HOST_NVIDIA_DIR.
• --host-mig-manager-state-file value, -o value — host path where the host's systemd mig-manager state file is located. default: "/etc/systemd/system/nvidia-mig-manager.service.d/override.conf". env var: $HOST_MIG_MANAGER_STATE_FILE. yes that path is a whole journey, she contains multitudes 💜
• --host-kubelet-systemd-service value, -k value — name of the host's 'kubelet' systemd service which may need to be shutdown/restarted across a MIG mode reconfiguration. default: "kubelet.service". env var: $HOST_KUBELET_SYSTEMD_SERVICE.
• --gpu-clients-file value, -g value — the path to the file listing the GPU clients that need to be shutdown across a MIG configuration. env var: $GPU_CLIENTS_FILE.
• --with-reboot, -r — reboot the node if changing the MIG mode fails for any reason. default: false. env var: $WITH_REBOOT. that's the "if it all goes sideways, just send the whole node to its room" button 🔥
• --with-shutdown-host-gpu-clients, -d — shutdown/restart any required host GPU clients across a MIG configuration. default: false. env var: $WITH_SHUTDOWN_HOST_GPU_CLIENTS.
• --default-gpu-clients-namespace value, -p value — default name of the Kubernetes namespace in which the GPU client Pods are installed in. default: "default". env var: $DEFAULT_GPU_CLIENTS_NAMESPACE.
• --nvidia-driver-root value, --driver-root value, -t value — root path to the NVIDIA driver installation. only used if --cdi-enabled is set. default: "/run/nvidia/driver". env vars: $NVIDIA_DRIVER_ROOT, $DRIVER_ROOT. three names, one queen.
• --driver-root-ctr-path value, -a value — root path to the NVIDIA driver installation mounted in the container. only used if --cdi-enabled is set. default: "/run/nvidia/driver". env var: $DRIVER_ROOT_CTR_PATH.
• --cdi-enabled — enable CDI support. default: false. env var: $CDI_ENABLED. the little switch that wakes up a whole bunch of the options above, btw.
• --dev-root value, -b value — root path to the NVIDIA device nodes. only used if --cdi-enabled is set. env var: $NVIDIA_DEV_ROOT. (note: no default listed here, bestie, don't invent one.)
• --dev-root-ctr-path value, -j value — root path to the NVIDIA device nodes mounted in the container. only used if --cdi-enabled is set. env var: $DEV_ROOT_CTR_PATH.
• --nvidia-cdi-hook-path value — path to nvidia-cdi-hook binary on the host. default: "/usr/local/nvidia/toolkit/nvidia-cdi-hook". env var: $NVIDIA_CDI_HOOK_PATH.
• --help, -h — show help. your bestie when you're lost.
• --version, -v — print the version. (ironic given the version says unknown, but ok girl, we support you.)

💀 heads up: options marked "only used if --cdi-enabled is set" genuinely just sit there doing nothing until you flip --cdi-enabled. no notes, that's the design.

---

💖 SEE ALSO 💖

the full documentation for nvidia-mig-manager is maintained as a Texinfo manual. if the info and nvidia-mig-manager programs are properly installed at your site, this command

info nvidia-mig-manager

should give you access to the complete manual. so if you want the whole lore, that's where the whole lore lives 🧚

---

and that's the page, bestie. this was written for nvidia-mig-manager 0.14.2, dated April 2026. go forth and reconfigure some MIGs, you absolute icon 🎀✨ slay, but like, safely — periodt.