# NVIDIA Aggressive Fan Control Service

## What this is

A headless fan control daemon for the **NVIDIA RTX Pro 6000 Blackwell Workstation Edition**
running on Ubuntu with driver 590. No X11 or display manager required.

### Why it exists

The RTX Pro 6000's default fan curve is tuned for quiet office/workstation use.
Under sustained rendering workloads it kept fans at ~50% even at 83°C with only 10°C
of thermal headroom, causing hard power-off crashes.

This daemon replaces the stock curve with an aggressive gaming-style curve via the
NVML Python API (pynvml), keeping the GPU in the 70-74°C range at 83-86% fan speed
instead of crashing.

### Results (tested 2026-02-17)

| Metric          | Stock curve (crashed) | This daemon (stable) |
|-----------------|-----------------------|----------------------|
| GPU peak temp   | 83°C                  | 73°C                 |
| Thermal headroom| 10°C                  | 20°C                 |
| Fan at load     | 50%                   | 86%                  |
| 32-image render | Crash                 | Completed            |

---

## Fan curve

| GPU Temp | Fan Speed |
|----------|-----------|
| 0°C      | 30%       |
| 45°C     | 40%       |
| 55°C     | 55%       |
| 63°C     | 70%       |
| 70°C     | 82%       |
| 75°C     | 90%       |
| 80°C+    | 100%      |

Values between points are interpolated linearly.

---

## Files

- `nvidia-fan-control.py` — the daemon script
- `nvidia-fan-control.service` — systemd unit file
- `README.md` — this file

---

## Prerequisites

```bash
# pynvml must be installed
pip3 show nvidia-ml-py

# If missing:
pip3 install nvidia-ml-py
```

---

## Install / Reinstall steps

### 1. Copy the daemon script
```bash
sudo cp nvidia-fan-control.py /usr/local/bin/nvidia-fan-control.py
sudo chmod +x /usr/local/bin/nvidia-fan-control.py
```

### 2. Test it manually first (Ctrl+C to stop, auto-restores fan control)
```bash
sudo python3 /usr/local/bin/nvidia-fan-control.py
```

### 3. Install the systemd service
```bash
sudo cp nvidia-fan-control.service /etc/systemd/system/nvidia-fan-control.service
sudo systemctl daemon-reload
sudo systemctl enable nvidia-fan-control.service
sudo systemctl start nvidia-fan-control.service
```

### 4. Verify it's running
```bash
sudo systemctl status nvidia-fan-control.service
```

---

## Useful commands

```bash
# Watch live fan log
sudo journalctl -u nvidia-fan-control.service -f

# Stop the service (fans revert to automatic)
sudo systemctl stop nvidia-fan-control.service

# Disable from autostart
sudo systemctl disable nvidia-fan-control.service

# Check GPU temps and fan live
watch -n 1 nvidia-smi --query-gpu=temperature.gpu,fan.speed,power.draw,power.limit --format=csv,noheader
```

---

## Notes

- Tested on: Ubuntu 24.04, Driver 590.48.01, RTX Pro 6000 Blackwell
- Power limit does NOT need to be reduced (600W default is fine with proper cooling)
- The daemon runs as root (required for NVML fan control)
- On daemon exit (including service stop), fan control is automatically returned to the driver