Repositioned from a GPU monitor into a home-lab monitoring essential — GPU & local-AI first, whole-host at a glance.
Highlights
- 🧠 Model-level drill-down — see which model is loaded and its VRAM. Ollama validated (live per-model VRAM via
/api/ps); vLLM, HF TGI, llama.cpp, Automatic1111, ComfyUI detected best-effort. Pluggable probe registry. - 🚦 Contention intelligence — detects VRAM pressure, scans GPU containers' logs for OOM events, and correlates who lost to whom.
- 💡 Plain-language recommendations in the dashboard.
- 🖥️ Whole-host health — CPU, RAM, load, uptime, temperature, disk usage.
- 📈 History downsampled on read — any time range stays fast and readable.
- 🔍 Still zero-config: every GPU process is auto-attributed to its Docker container by name.
Run
git clone https://github.com/SikamikanikoBG/nvidia-gpu-monitor.git
cd nvidia-gpu-monitor && docker compose up -d --buildOpen http://:9800 on your LAN/VPN.