utkuozdemir/nvidia_gpu_exporter

What it solves

It provides a way to monitor NVIDIA GPUs across different operating systems and environments. It is specifically designed for scenarios where heavy datacenter tooling (like the NVIDIA GPU Operator) is overkill or unavailable, such as on consumer GeForce/RTX cards, small Kubernetes clusters, edge boxes, homelabs, or gaming rigs.

How it works

By default, the exporter uses the nvidia-smi binary to collect, parse, and export metrics to Prometheus. It can also be configured to run nvidia-smi remotely. For Linux users, an experimental native NVML backend is available that reads metrics directly from the NVIDIA driver library, providing additional data like PCIe throughput and XID error counters.

Who it’s for

• System administrators and developers managing mixed fleets of GPUs.
• Homelab enthusiasts and gamers who want to track GPU stats on a dashboard.
• Users in restricted or virtualized environments (vGPU guests, MIG slices) where deep counters are not exposed.

Highlights

• Cross-platform support for Linux, Windows, and macOS.
• Works on bare metal, in Docker, or on Kubernetes.
• Auto-discovery of metric fields for future compatibility.
• Optional per-process GPU memory tracking.
• Includes pre-configured Grafana dashboards for detailed per-GPU and multi-GPU overview views.