High-Frequency Energy Monitoring for GPU Nodes

A comprehensive toolkit for high-precision energy monitoring of GPU compute nodes. IdleWatts provides microsecond-level precision to capture both idle and active power states, supporting energy efficiency research, optimization, and benchmarking in high-performance computing environments.

🚀 Features

• High-frequency sampling (up to 100Hz) of CPU power using Intel RAPL
• Detailed GPU power monitoring via NVIDIA SMI with support for multiple GPUs
• System-level power monitoring through iDRAC/Redfish API
• Combined monitoring for synchronized data collection across components
• Customizable sampling intervals and duration
• Data export to CSV for further analysis
• Visualization tools for power consumption patterns
• Support for both idle and load power profiling

📋 Components

The toolkit includes multiple modules:

Idle Power Profiling

• idle_cpu_high_freq.py: CPU power monitoring using Intel RAPL
• idle_gpu_high_freq.py: GPU power monitoring using NVIDIA-SMI
• idle_system_high_freq.py: System power monitoring using iDRAC/Redfish
• run_energy_monitoring.sh: Combined monitoring script

Load Power Profiling

• Tools for measuring energy consumption under various workloads
• Benchmark suites for standardized comparisons
• Application-specific power profiling

Analysis Tools

• Scripts for data processing and visualization
• Statistical analysis of power patterns
• Component-level energy breakdown

🔧 Requirements

• Python 3.6+
• For CPU monitoring:
  • Intel CPU with RAPL support
  • Root access
• For GPU monitoring:
  • NVIDIA GPU
  • NVIDIA drivers with nvidia-smi
  • nvidia-ml-py3 Python package
• For system monitoring:
  • Dell server with iDRAC
  • Network access to iDRAC interface
  • requests Python package

📦 Installation

# Clone the repository
git clone https://github.com/yourusername/idlewatts.git
cd idlewatts

# Set up virtual environment
python -m venv energy_monitor_env
source energy_monitor_env/bin/activate

# Install dependencies
pip install -r requirements.txt

# Make scripts executable
chmod +x *.py *.sh

💻 Usage

Idle Power Monitoring

# Run the combined monitoring script
./run_energy_monitoring.sh [DURATION_SECONDS] [INTERVAL_SECONDS]

# Examples:
# Monitor for 1 hour with 10ms sampling
./run_energy_monitoring.sh 3600 0.01

# Monitor indefinitely until manually stopped
./run_energy_monitoring.sh 0 0.01

Individual Component Monitoring

# CPU monitoring only
sudo python idle_cpu_high_freq.py -i 0.01 -o cpu_data.csv

# GPU monitoring only
python idle_gpu_high_freq.py -i 0.01 -o gpu_data.csv

# System monitoring (if iDRAC is accessible)
python idle_system_high_freq.py --host IDRAC_IP --username USER --password PASS -i 0.1

Analyzing Results

# Analyze collected data
python analyze_energy_data.py --cpu-data cpu_data.csv --gpu-data gpu_data.csv

# Generate visualizations
python visualize_power.py --data-dir ./energy_data_YYYYMMDD_HHMMSS/

🔍 Use Cases

• Baseline energy benchmarking for GPU nodes
• Identification of power optimization opportunities
• Energy efficiency research in HPC environments
• Supporting green computing initiatives
• Validating power management policies
• Component-level power attribution

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

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

🙏 Acknowledgements

• Intel for RAPL technology
• NVIDIA for power monitoring capabilities
• Dell for iDRAC API documentation