Baram: Advanced System Cooling Management Tool

A Linux utility designed to manage fans on datacenter GPUs (like Tesla P40) mounted in ATX cases. Named after the Korean word for wind (바람), Baram provides intelligent, adaptive fan control based on GPU temperature and power consumption.

Overview

Baram is a specialized cooling management tool developed for NVIDIA datacenter GPUs repurposed in consumer ATX cases. It monitors GPU temperature and power via NVML and controls motherboard fan headers via the Linux hwmon interface.

Key Features

• Automatic hwmon Detection: Automatically finds compatible fan control interfaces
• Dynamic Temperature Control: Adaptive fan curves based on real-time GPU temperature
• Power Spike Detection: Proactive cooling boost when GPU power consumption spikes
• Graceful Shutdown: Proper signal handling and resource cleanup
• Configurable: Full configuration via file or command line
• Systemd Integration: Production-ready service file included
• CSV Logging: Detailed telemetry logging for analysis

Inspiration

Why does this project exits? I found a cheap Nvidia P40 on ebay and all I had was an old AM4 gammer system to test it our on. So I threw it in the spare 8x slot and figured out a way to cool it while keeping teh noise at a reasonable level.

The solution: Position the GPU's intake directly in front of the ATX case's intake fans, providing fresh air directly to the GPU rather than pre-warmed air from CPU cooling. This not only enhanced cooling efficiency but also significantly reduced noise levels.

Requirements

• Python 3.8+
• NVIDIA GPU with driver installed
• Linux with hwmon support (most motherboards)
• Root privileges (for hwmon access)

Python Dependencies

pip install pynvml

Installation

Quick Start

# Clone the repository
git clone https://github.com/r3d91ll/Baram.git
cd Baram

# Install dependencies
pip install -r requirements.txt

# List available hwmon devices
sudo python baram.py --list-hwmon

# Run with auto-detection
sudo python baram.py

System Installation

# Install to /opt/baram
sudo mkdir -p /opt/baram /etc/baram /var/log/baram
sudo cp baram.py /opt/baram/
sudo cp baram.conf /etc/baram/

# Install systemd service
sudo cp baram.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now baram

Usage

Command Line

# Show help
python baram.py --help

# List available hwmon devices and PWM channels
sudo python baram.py --list-hwmon

# Run with defaults (auto-detect hwmon)
sudo python baram.py

# Specify GPU index and hwmon device
sudo python baram.py --gpu-index 0 --hwmon-device hwmon2 --pwm-channel 1

# Override temperature thresholds
sudo python baram.py --min-temp 35 --max-temp 75

Command Line Options

Option	Description	Default
--config	Path to configuration file	/etc/baram/baram.conf
--gpu-index	GPU index to monitor (0-based)	0
--hwmon-device	Hwmon device path/name (empty = auto)	Auto-detect
--pwm-channel	PWM channel number (1-based)	1
--min-temp	Minimum temperature threshold (°C)	40
--max-temp	Maximum temperature threshold (°C)	80
--min-pwm-value	Minimum PWM value (0-255)	20
--max-pwm-value	Maximum PWM value (0-255)	255
--list-hwmon	List available hwmon devices and exit	-

Configuration File

Copy baram.conf to /etc/baram/baram.conf and edit as needed. Command line arguments override configuration file settings.

See baram.conf for all available options with detailed comments.

Systemd Service

# Start the service
sudo systemctl start baram

# Check status
sudo systemctl status baram

# View logs
sudo journalctl -u baram -f

# Enable at boot
sudo systemctl enable baram

How It Works

Temperature-Based Fan Curve

Baram uses a linear interpolation between temperature thresholds:

GPU Temp	PWM Range
< 30°C	0-0
30-40°C	0-30
40-50°C	30-60
50-60°C	60-90
60-70°C	90-120
70-max°C	120-255

Power Spike Detection

When GPU power consumption exceeds the threshold for consecutive readings:

1. Spike counter increments
2. After wattage_spike_count spikes, minimum PWM is raised to wattage_pwm_value
3. When power drops, counter decrements
4. When counter reaches 0, minimum PWM returns to normal

This provides proactive cooling before temperature rises during heavy GPU loads.

Hardware Setup

Tested Configuration

• Motherboard: ASUS ROG Strix B550-XE Gaming (AMD chipset)
• GPU: NVIDIA Tesla P40 24GB
• Fan Shroud: 3D Printed Fan Shroud
• Fans: ARCTIC S4028-15K (40x40x28mm, 1400-15000 RPM, PWM)

Finding Your hwmon Device

# List all hwmon devices
ls -la /sys/class/hwmon/

# Check device names
for d in /sys/class/hwmon/hwmon*; do echo "$d: $(cat $d/name 2>/dev/null)"; done

# Or use Baram's built-in detection
sudo python baram.py --list-hwmon

Common motherboard sensor chips:

• Nuvoton: nct6775, nct6776, nct6779, nct6791-nct6798
• ITE: it8720, it8721, it8728, it8771, it8772
• Winbond: w83627, w83667, w83795
• ASUS: asus_wmi_sensors, asus-ec-sensors

Logging

Baram creates two log files in /var/log/baram/:

• baram.log - Application logs (startup, errors, debug info)
• baram.csv - Telemetry data (timestamp, temp, RPM, PWM, power)

Troubleshooting

Permission Denied

Hwmon files require root access:

sudo python baram.py

No hwmon Device Found

1. Check if your motherboard's sensor driver is loaded:

   lsmod | grep -E "nct6775|it87|w83627"

2. Try loading the driver manually:

   sudo modprobe nct6775  # or it87, etc.

3. Add to /etc/modules for persistence

Wrong Fan Responding

Use --list-hwmon to identify the correct device and channel, then specify explicitly:

sudo python baram.py --hwmon-device /sys/class/hwmon/hwmon3 --pwm-channel 2

GPU Load Testing

For testing cooling performance under load:

• gpu-burn - GPU stress test

Compatibility

Verified Working

• ASUS ROG Strix B550-XE Gaming (AMD, nct6775)

Should Work

• Most motherboards with Linux hwmon support
• Any NVIDIA GPU supported by NVML

Known Limitations

• hwmon paths (/sys/class/hwmon/hwmonN) can change between boots
• Some motherboards require specific kernel modules
• Virtual machines may not expose hwmon interfaces

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

License

This project is licensed under the GPL-3.0 License - see the LICENSE file for details.