NVIDIA-TUNER-CPP

A simple C++ CLI tool for overclocking, undervolting and controlling the fan of NVIDIA GPUs on Linux. Using the NVML library it equally supports X11 and Wayland.

Features

• Set core clock offset.
• Set memory clock offset.
• Set maximum boost core clock.
• Set maximum boost memory clock.
• Set power limit.
• PI-based temperature control for automatic fan management.
• Automatically set the fan control back to default on termination.

Usage

This tool is still under testing and it is impossible for me to guarantee that it works on every hardware, so use it at your own risk

Show all possible options:

./nvidia-tuner --help

Usage example with overclocking:

./nvidia-tuner --core-clock-offset 150 --memory-clock-offset 800 --power-limit 180

Usage example with PI temperature control:

./nvidia-tuner --target-temperature 70 --proportional-gain 2.5 --integral-gain 0.15

Combined example:

./nvidia-tuner --core-clock-offset 150 --memory-clock-offset 800 --power-limit 180 --target-temperature 70

The PI controller automatically adjusts fan speed to maintain the target temperature. The proportional and integral gains can be tuned for different response characteristics - higher proportional gain gives faster response, while integral gain eliminates steady-state error.

Compilation

To compile from source, you'll need the NVIDIA ML development library and CMake:

# Install dependencies
sudo apt install libnvidia-ml-dev cmake build-essential

# Clone and build
git clone https://github.com/jukofyork/nvidia-tuner-cpp
cd nvidia-tuner-cpp
mkdir build
cd build
cmake ..
make -j$(nproc)

The compiled binary will be located at build/nvidia-tuner.

Run on startup

1. Copy the binary to /usr/local/sbin/.
2. Create the systemd service file /etc/systemd/system/nvidia-tuner.service with the following content:

[Unit]
Description=NVIDIA GPU Tuner and Temperature Controller
After=graphical.target

[Service]
Type=exec
ExecStart=/usr/local/sbin/nvidia-tuner --core-clock-offset 150 --memory-clock-offset 800 --power-limit 180 --target-temperature 70
Restart=always
RestartSec=5
StandardOutput=journal
StandardError=journal

[Install]
WantedBy=graphical.target

3. Reload the systemd manager configuration to recognize the new service:

sudo systemctl daemon-reload

4. Start the service:

sudo systemctl start nvidia-tuner.service

5. Enable the service to start automatically at boot:

sudo systemctl enable nvidia-tuner.service

6. Check the systemd journal for any errors:

sudo journalctl -u nvidia-tuner.service

PI Controller Tuning

The PI controller uses two main parameters:

• Proportional Gain (-p): Controls immediate response to temperature changes. Higher values = faster response but may cause oscillation.
• Integral Gain (-i): Eliminates steady-state temperature error. Higher values = better accuracy but may cause instability.

Default values (4.0 and 0.2) work well for most GPUs, but you can experiment with different values if needed.