Skip to main content

nvidia-smi Command Reference

nvidia-smi is the NVIDIA System Management Interface, a command-line utility for monitoring and managing NVIDIA GPU devices. Below is an overview and explanation of the command-line options.

Basic Options​

  • -h, --help: Display help information and exit.

List Options​

  • -L, --list-gpus: Display the list of GPUs connected to the system.
  • -B, --list-excluded-gpus: Display the list of excluded GPUs in the system.

Summary Options​

  • <no arguments>: Display a summary of GPUs connected to the system.
  • -i, --id=: Specify a particular GPU.
  • -f, --filename=: Log to a specified file instead of standard output.
  • -l, --loop=: Probe at specified second intervals until Ctrl+C.

Query Options​

  • -q, --query: Display GPU or unit information.
  • -u, --unit: Display unit attributes instead of GPU attributes.
  • -x, --xml-format: Generate XML output.
  • --dtd: Embed DTD when displaying XML output.
  • -d, --display=: Display only selected information (e.g., memory, utilization, etc.).
  • -lms, --loop-ms=: Probe at specified millisecond intervals until Ctrl+C.

Selective Query Options​

  • --query-gpu: Query GPU information.
  • --query-supported-clocks: List supported clocks.
  • --query-compute-apps: List currently active compute processes.
  • --query-accounted-apps: List accounted compute processes.
  • --query-retired-pages: List retired device memory pages.
  • --query-remapped-rows: Query information about remapped rows.
  • --format=: Specify format options, such as csv.

Device Modification Options​

  • -pm, --persistence-mode=: Set persistence mode.
  • -e, --ecc-config=: Toggle ECC support.
  • -p, --reset-ecc-errors=: Reset ECC error counts.
  • -c, --compute-mode=: Set the mode for compute applications.
  • --gom=: Set the GPU operation mode.
  • -r, --gpu-reset: Trigger a GPU reset.
  • -vm, --virt-mode=: Toggle GPU virtualization mode.
  • -lgc, --lock-gpu-clocks=: Lock GPU clock speeds.
  • -rgc, --reset-gpu-clocks: Reset GPU clocks to default values.
  • -lmc, --lock-memory-clocks=: Lock memory clock speeds.
  • -rmc, --reset-memory-clocks: Reset memory clocks to default values.
  • -ac, --applications-clocks=: Specify GPU clock speeds for application runtime.
  • -pl, --power-limit=: Specify the maximum power management limit.
  • -am, --accounting-mode=: Enable or disable accounting mode.
  • -mig, --multi-instance-gpu=: Enable or disable Multi-Instance GPU.
  • -gtt, --gpu-target-temp=: Set the GPU target temperature.

Unit Modification Options​

  • -t, --toggle-led=: Set the unit LED state.

Display DTD Options​

  • --dtd: Print the device DTD and exit.

Statistics Options​

  • stats: Display device statistics (deprecated).

Device Monitoring​

  • dmon: Display device statistics in a scrolling format.
  • daemon: Run as a daemon in the background to monitor devices.
  • replay: Used to replay/extract persistent statistics generated by the daemon.

Process Monitoring​

  • pmon: Display process statistics in a scrolling format.

Other Features​

  • topo: Display device/system topology.
  • drain: Display/modify GPU drain state for power idle.
  • nvlink: Display device NVLINK information.
  • clocks: Control and query clock information.
  • encodersessions: Display device encoder session information.
  • fbcsessions: Display device FBC session information.
  • vgpu: Display vGPU information.
  • mig: Provide MIG management controls.
  • compute-policy: Control and query compute policies.
  • boost-slider: Control and query boost sliders.
  • power-hint: Estimate GPU power usage.
  • base-clocks: Query GPU base clocks.
  • ccu: Control and query the Counter Collection Unit.

These options provide deep monitoring and management capabilities for NVIDIA GPUs, from simple status queries to complex configuration adjustments. When using these options, some may require administrator privileges or specific hardware support.