nvidia-smi

command-line utility for monitoring and managing NVIDIA GPU devices

TLDR

Show GPU status

$ nvidia-smi

Show detailed GPU info

$ nvidia-smi -q

Monitor GPU continuously

$ nvidia-smi -l [1]

Show only specific GPU

$ nvidia-smi -i [0]

Show processes using GPU

$ nvidia-smi pmon

Query specific attributes

$ nvidia-smi --query-gpu=name,memory.total,memory.used --format=csv

Set power limit (watts)

$ nvidia-smi -pl [250]

Set persistence mode

$ nvidia-smi -pm [1]

Reset GPU

$ nvidia-smi -r

Show driver version

$ nvidia-smi --query-gpu=driver_version --format=csv,noheader

nvidia-smi (NVIDIA System Management Interface) is a command-line utility for monitoring and managing NVIDIA GPU devices. It provides information about GPU utilization, memory usage, temperature, power consumption, and running processes.The tool can display real-time statistics, query specific GPU attributes, configure power and clock settings, and manage compute processes. It's essential for GPU monitoring in machine learning, rendering, and scientific computing environments.nvidia-smi is included with the NVIDIA driver package and works with Tesla, Quadro, and GeForce GPUs (feature availability varies).

PARAMETERS

-l sec, --loop= sec

Continuously update every sec seconds.

-i id

Target specific GPU by index.

-q, --query

Display detailed information.

-d type

Display specific info type (MEMORY, UTILIZATION, TEMPERATURE, POWER, CLOCK).

--query-gpu= attrs

Query specific attributes.

--format= format

Output format (csv, noheader, nounits).

-pm mode

Set persistence mode (0 or 1).

-pl watts

Set power limit.

-r, --gpu-reset

Reset GPU.

pmon

Process monitoring mode.

topo

Show GPU topology.

nvlink

Show NVLink status.

QUERY ATTRIBUTES

name: GPU namememory.total: Total memorymemory.used: Used memorymemory.free: Free memoryutilization.gpu: GPU utilization %utilization.memory: Memory utilization %temperature.gpu: GPU temperaturepower.draw: Power consumption

CAVEATS

Requires NVIDIA proprietary drivers. Some features limited on consumer GPUs. Power management requires root. Persistence mode recommended for servers to avoid initialization delays.

HISTORY

nvidia-smi has been part of the NVIDIA driver package since the early 2000s. It evolved alongside NVIDIA's GPU computing platforms (CUDA), gaining features for monitoring compute workloads. The tool became essential with the rise of GPU computing for machine learning and scientific applications.