A powerful command-line utility for monitoring and controlling NVIDIA GPUs using the NVML library. Built with a focus on simplicity, performance, and reliability.
# Build from source
git clone https://github.com/xl0/nvml-tool
cd nvml-tool
make
sudo make install
# Or install to custom location
make install PREFIX=/usr/local# Show information for all GPUs
nvml-tool info
# Monitor specific GPU
nvml-tool info -d 0
# Get JSON output for automation
nvml-tool info json
# Monitor power consumption
nvml-tool power
# Set power limit (requires root)
sudo nvml-tool power set 250 -d 0
# Control fan speed (requires root)
sudo nvml-tool fan set 80 -d 0
# Restore automatic fan control
sudo nvml-tool fan restore -d 0
# Dynamic fan control with temperature setpoints (requires root)
sudo nvml-tool fanctl 50:30 70:60 80:90 -d 0
# Quick status overview
nvml-tool statusDisplay comprehensive device information including name, UUID, temperature, memory usage, fan speed, and power consumption.
nvml-tool info # All devices, human-readable
nvml-tool info -d 0 # Device 0 only
nvml-tool info json # JSON output
nvml-tool info -d 0-2 json # Devices 0-2, JSON formatMonitor or control GPU power consumption and limits.
nvml-tool power # Show current power usage
nvml-tool power -d 0 # Power for device 0
sudo nvml-tool power set 200 -d 0 # Set 200W limit on device 0Control GPU fan speeds manually or restore automatic control.
nvml-tool fan # Show current fan speeds
sudo nvml-tool fan set 75 -d 0 # Set 75% fan speed on device 0
sudo nvml-tool fan restore # Restore automatic control (all devices)
sudo nvml-tool fan restore -d 0 # Restore automatic control (device 0)Display GPU temperatures in various units.
nvml-tool temp # Celsius (default)
nvml-tool temp --temp-unit F # Fahrenheit
nvml-tool temp --temp-unit K # KelvinShow compact status overview with temperature, fan speed, and power.
nvml-tool status # All devices
nvml-tool status -d 0-1 # Devices 0 and 1Dynamic fan control using temperature setpoints with linear interpolation. Continuously monitors GPU temperature and adjusts fan speed based on the defined temperature-to-fan-speed mapping.
Requirements: Root access, controllable fans
# Basic usage with temperature:fan% setpoints
sudo nvml-tool fanctl 50:30 70:60 80:90 -d 0
# Multiple setpoints for fine control
sudo nvml-tool fanctl 40:20 50:30 60:45 70:60 80:80 90:100
# Control all devices
sudo nvml-tool fanctl 50:30 70:60 80:90How it works:
- Takes temperature:fan-speed setpoints (e.g.,
70:60= 70°C → 60% fan speed) - Uses linear interpolation between setpoints for smooth transitions
- Updates fan speeds every 2 seconds based on current GPU temperature
- Shows live status updates when run in terminal
- Automatically restores automatic fan control on exit (Ctrl-C)
Safety considerations:
- Monitor temperatures carefully when using manual fan control
- Insufficient cooling can damage your GPU
- Use
Ctrl-Cto exit and restore automatic control - Fan control is reset to automatic if the tool exits unexpectedly
List all available GPUs with their IDs, UUIDs, and names.
nvml-tool list # Simple device listing-d 0 # Single device
-d 0-2 # Range (devices 0, 1, 2)
-d 0,2,4 # List (devices 0, 2, 4)-u GPU-abc123 # Partial UUID match
-u GPU-abc123-def456-789 # Full UUID--temp-unit C # Celsius (default)
--temp-unit F # Fahrenheit
--temp-unit K # KelvinPerfect for automation and scripting:
nvml-tool info json | jq '.[0].power_usage_watts'
nvml-tool status | awk -F: '{print $1 ": " $2}' | column -t- GCC or compatible C compiler
- NVML library (from NVIDIA drivers, CUDA toolkit, or system packages)
- pkg-config
If build fails with NVML detection errors:
# Check if NVML is installed
pkg-config --list-all | grep nvidia-ml
# Pass CFLAGS and LIBS manually if pkg-config is wrong
make NVML_CFLAGS="-I/usr/local/cuda/include" NVML_LIBS="-L/usr/local/cuda/lib64 -lnvidia-ml"
Most monitoring commands work as regular user, but control commands require root:
# Monitoring (no root required)
nvml-tool info
nvml-tool power
nvml-tool temp
# Control (requires root)
sudo nvml-tool power set 200 -d 0
sudo nvml-tool fan set 75 -d 0
sudo nvml-tool fanctl 50:30 70:60 80:90 -d 0Error: "Device has no controllable fans"
- Some GPUs don't support manual fan control
- Older NVIDIA drivers may not support fan control
- Check if your GPU model supports fan control
Fanctl not working as expected:
- Ensure you're running as root (
sudo) - Check that NVML version supports fan control policies
- Monitor GPU temperatures to verify setpoints are reasonable
=== Device 0: NVIDIA RTX 4090 ===
UUID: GPU-12345678-abcd-ef12-3456-789abcdef012
Temperature: 45.2°C
Memory: 1024 MB / 24576 MB (4.2%)
Fan Speed: 35%
Power: 125.5W / 450.0W (27.9%)
[
{
"device_id": 0,
"name": "NVIDIA RTX 4090",
"uuid": "GPU-12345678-abcd-ef12-3456-789abcdef012",
"temperature": 45.2,
"temperature_unit": "C",
"memory_total_mb": 24576,
"memory_used_mb": 1024,
"memory_free_mb": 23552,
"fan_speed_percent": 35,
"power_usage_watts": 125.50,
"power_limit_watts": 450.00
}
]0:45.2C,35%,125.5W
1:42.1C,40%,98.2W
2:50.3C,45%,156.7W
Starting dynamic fan control for 1 device(s) (Ctrl-C to exit)
Setpoints: 50:30% 70:60% 80:90%
0:52.3C -> 42%
0:53.1C -> 44%
0:54.2C -> 47%
^C
Restoring automatic fan control...