laperf-power 0.1.3

Lightweight real-time power and resource monitoring tool for AI workloads

LaPerf Power

Lightweight real-time power and resource monitoring tool for AI/ML workloads.

Features

• Real-time monitoring with live console output
• Cross-platform support: macOS (Apple Silicon + Intel), Linux (NVIDIA), Windows
• Comprehensive metrics: GPU/CPU power, utilization, VRAM, temperature, RAM, battery
• Statistical analysis: P50/P95 percentiles for all metrics
• JSON export for further analysis
• Zero heavy dependencies: Only requires psutil (~5 MB installed)

Installation

Quick run without installation (recommended)

uvx laperf-power

Install as a global tool

# Using uv
uv tool install laperf-power

# Using pip
pip install laperf-power

Usage

# Start monitoring with default settings (10s interval)
laperf-power

# Custom sampling interval (faster sampling = more overhead)
laperf-power --interval 1.0

# Without sudo (basic metrics only on macOS)
laperf-power --no-sudo

# Save results to JSON
laperf-power --output power_metrics.json

Press Ctrl+C to stop and view statistics.

What It Monitors

GPU Metrics

• Power consumption (Watts) - NVIDIA via nvidia-smi, macOS via powermetrics
• Utilization (%) - GPU compute usage
• VRAM (GB) - Memory used/total
• Temperature (°C) - Die temperature

CPU Metrics

• Power consumption (Watts) - macOS only with sudo
• Utilization (%) - Average across all cores

System Metrics

• RAM usage (GB) - Process memory consumption
• Battery drain (%) - Change during monitoring period

Platform Support

Platform	GPU Power	GPU Stats	CPU Power	Notes
macOS (Apple Silicon)	✅ (with sudo)	✅	✅ (with sudo)	Via powermetrics and ioreg
macOS (Intel)	❌	❌	✅ (with sudo)	Via powermetrics
Linux (NVIDIA)	✅	✅	❌	Via nvidia-smi
Linux (AMD/Intel)	❌	❌	❌	CPU/RAM only
Windows	❌	❌	❌	CPU/RAM only

Example Output

Real-time Monitoring

================================================================================
REAL-TIME POWER MONITORING
================================================================================
Started: 2025-11-27 14:30:00
Interval: 10.0s
================================================================================

Press Ctrl+C to stop and view statistics

[Sample #42] GPU: 11.7W 32% 8.2GB | CPU: 15% 1.0W | RAM: 16.3GB | Temp: 45°C

Final Statistics

================================================================================
MONITORING SUMMARY
================================================================================

Duration: 420.0s
Samples collected: 42

Metric             |     P50 |     P95
-----------------------------------------
GPU Power          | 11.70W  | 13.20W
CPU Power          |  1.00W  |  1.50W
GPU Utilization    |    32%  |    45%
GPU VRAM           |  8.20GB |  8.50GB
CPU Utilization    |    15%  |    22%
RAM Usage          | 16.30GB | 16.70GB
GPU Temperature    |    45°C |    48°C

Battery: 85.0% → 83.5% (drain: 1.5%)

================================================================================

macOS sudo Setup (Optional)

For detailed GPU/CPU power metrics on macOS, laperf-power uses sudo powermetrics.

Option 1: Enter password when prompted (recommended for occasional use)

Option 2: Passwordless sudo (for frequent use)

Add to /etc/sudoers (use sudo visudo):

your_username ALL=(ALL) NOPASSWD: /usr/bin/powermetrics

Use Cases

• AI/ML Development: Monitor power usage during model training/inference
• Hardware Evaluation: Compare power efficiency across different GPUs
• Performance Optimization: Identify power/performance bottlenecks
• Battery Life Testing: Track power consumption on laptops
• System Monitoring: Real-time resource usage dashboard

Part of LaPerf

laperf-power is extracted from LaPerf - a comprehensive AI hardware benchmark suite. For full benchmarking capabilities (embeddings, LLMs, VLMs), check out the main project.

License

Apache-2.0

Contributing

Issues and PRs welcome at https://github.com/bogdanminko/laperf