// the find
utkuozdemir/nvidia_gpu_exporter
Nvidia GPU exporter for prometheus using nvidia-smi binary
A Prometheus exporter for Nvidia GPU metrics that wraps nvidia-smi instead of using NVML C bindings. Works on Windows, Linux, and macOS wherever nvidia-smi is available. Good fit for developers running local LLM inference or gaming rigs who want GPU metrics without standing up DCGM.
Auto-discovers whatever metrics nvidia-smi can expose on your driver version, so it stays forward-compatible without code changes. Ships a single static Go binary — no runtime dependencies, no Docker required, works as a Windows service. Includes a ready-made Grafana dashboard (ID 14574) so you get a usable visualization immediately. Can run remotely and exec nvidia-smi over SSH, which is handy for headless inference boxes.
The maintainer has openly flagged it as slow-to-respond for over a year — PRs and issues sit. Parsing nvidia-smi CSV output is inherently brittle; driver updates that change field names or formatting can silently break metrics. No multi-GPU aggregation metrics out of the box — you get per-GPU rows but no fleet-level rollups. The test coverage is thin and based on fixture files from 2021-era driver output, so newer GPU generations are untested territory.