nvidia_gpu_exporter

Why we included this project

When your GPU boxes sit there doing work you can't see, this exporter is what makes them legible. It's a small Prometheus exporter that pulls utilization, memory, power, and temperature off NVIDIA cards and exposes them as scrapeable metrics, with official Grafana dashboards for both single-GPU and multi-GPU views. The practical trick is that it leans on the nvidia-smi binary instead of deep driver bindings, so it runs on Windows, consumer GeForce/RTX cards, vGPU guests, and locked-down containers where datacenter tooling refuses to go. That makes it a natural pick for homelabs and small clusters that would rather skip the heavyweight NVIDIA GPU Operator, and for mixed fleets of old and new cards that need one consistent behavior everywhere. An experimental NVML backend on Linux adds MIG-instance metrics, XID errors, energy counters, and PCIe throughput without renaming any existing metrics, so dashboards you already have keep working. If you're on datacenter cards with the GPU Operator already in place, DCGM-exporter.