🇺🇸 English | 🇩🇪 Deutsch | 🇷🇺 Русский | 🇫🇷 Français | 🇪🇸 Español | 🇯🇵 日本語 | 🇨🇳 中文 | 🇵🇹 Português | 🇰🇷 한국어 | 🇮🇳 हिंदी

MyGPU: Lightweight GPU Management Utility: a compact nvidia-smi wrapper with an elegant web dashboard.

Gallery

Web Dashboard

CLI

Why use this?

• Lightweight: Minimal resource footprint.
• Flexible: Runs as a CLI tool, or a full-featured Web Dashboard.
• Admin-Centric: Includes features like VRAM Enforcement (auto-kill processes exceeding limits) and Watchlists.
• Developer-Friendly: Built-in benchmarking and stress-testing tools (GEMM, Particle Physics) to validate system stability.

---

Features

• Real-time Monitoring:

  • Detailed GPU metrics (Utilization, VRAM, Power, Temp).
  • System metrics (CPU, RAM, etc.).

• Admin & Enforcement:

  • VRAM Caps: Set hard limits on VRAM usage per GPU.
  • Auto-Termination: Automatically terminate processes that violate VRAM policies (Admin only).
  • Watchlists: Monitor specific PIDs or process names.

• Benchmarking & Simulation:

  • Stress Testing: Configurable GEMM workloads to test thermal throttling and stability.
  • Visual Simulation: Interactive 3D particle physics simulation to visualize GPU load.

---

Roadmap & Future Work

Contributions are welcome! Main future points to cover would be:

• Multi-GPU Support: Enhanced handling for multi-card setups and NVLink topologies.
• Containerization: Official Docker support for easy deployment in containerized environments.
• Remote Access: SSH tunneling integration and secure remote management.
• Cross-Platform:
  • Linux Support (Ubuntu/Debian focus).
  • macOS Support (Apple Silicon monitoring).
• Hardware Agnostic:
  • AMD ROCm support.
  • Intel Arc support.
• Multi-Language Documentation: Supporting most popular GitHub languages.

See CONTRIBUTING.md for how to get involved.

---

Requirements

• OS: Windows 10/11, Linux, macOS
• Python: 3.10+
• Hardware: NVIDIA GPU (all platforms), Apple Silicon (macOS), or CPU-only.
• CUDA: Toolkit 12.x (Recommended for Benchmarking/Simulation on NVIDIA).
  • Note: If CUDA/MPS is not detected, some benchmarking features may be disabled.

---

Installation

The tool supports modular installation to fit your needs:

1. Minimal (CLI Only)

Best for headless servers or background monitoring.

• Command-line interface.
• Basic system/GPU metrics.

2. Standard (CLI + Web UI)

Best for most users.

• Includes Web Dashboard.
• REST API endpoints.
• Real-time charts.
• But no Simulation or benchmarking.

3. Full (Standard + Visualization)

Best for development and stress testing.

• Includes Simulation.
• PyTorch/CuPy dependencies for benchmarking.

Quick Start

1. Download or clone the repository.

2. Run Setup:

   Windows:

   .\setup.ps1

   Linux/macOS:

   chmod +x setup.sh
   ./setup.sh

3. Launch:

# Start the web dashboard (Standard/Full)
python health_monitor.py web

# Start the CLI
python health_monitor.py cli

---

License

See LICENSE for details.