A terminal-based utility written in Rust for monitoring NVIDIA GPU utilization via nvidia-smi with real-time graph visualization.

• Monitor multiple GPUs simultaneously
• Real-time display of GPU utilization, memory usage, temperature, and power consumption
• Cyberpunk/hacker-style visual theme with neon colors
• Data stored in memory only (no disk persistence)
• Data updates every second
• Stores last 60 minutes of metrics history

• Rust (version 1.75 or higher)
• NVIDIA drivers with nvidia-smi
• Linux (for nvidia-smi support)

src/
├── main.rs      # Entry point, terminal setup/cleanup, main loop
├── app.rs       # App struct - application state management
├── models.rs    # GpuDataPoint, GpuInfo - data models
├── nvidia.rs    # fetch_gpu_data - nvidia-smi integration
├── theme.rs     # Color constants (NEON_GREEN, DARK_BG, etc.)
└── ui.rs        # UI rendering functions

• models.rs — Data structures (Data Layer)
• nvidia.rs — External system integration (Infrastructure Layer)
• theme.rs — Style configuration (Presentation Constants)
• ui.rs — Interface rendering (Presentation Layer)
• app.rs — Application business logic (Application Layer)
• main.rs — Initialization and main loop (Entry Point)

cargo build --release

cargo run --release

Or after building:

./target/release/gpu-graph-cli

• q or Esc - exit the program

docker build -t gpu-graph-cli .

docker run --gpus all -it --rm gpu-graph-cli

Or use the convenience script:

./docker-run.sh

Important: nvidia-container-runtime is required for GPU access in Docker. Make sure it's installed on your system.

For each GPU, the following metrics are shown:

• GPU Utilization - percentage of GPU usage (progress bar + sparkline graph)
• Memory Usage - memory utilization (progress bar)
• Temperature - GPU temperature in °C (sparkline graph)
• Power Usage - power consumption in Watts

The interface features a cyberpunk aesthetic with:

• Neon green, cyan, magenta, and yellow color scheme
• Real-time sparkline graphs
• Status indicators that change color based on load levels

MIT