nvidia_gpu_exporter
Nvidia GPU exporter for prometheus using nvidia-smi binary OR using NVML
What is nvidia_gpu_exporter?
Nvidia GPU exporter for prometheus using nvidia-smi binary OR using NVML
Overview of nvidia_gpu_exporter
What it is
nvidia_gpu_exporter is a Prometheus exporter for NVIDIA GPUs. It collects metrics by reading directly from the NVIDIA driver library (NVML) or by running the nvidia-smi binary. It can attribute per-process GPU usage to containers, systemd services, and MIG instances. The exporter runs on Linux, Windows, and macOS, on bare metal, in Docker, or on Kubernetes.
Key features
- Collects utilization, memory, power, thermals, clocks, throttling reasons, ECC errors, encoder/decoder stats, and per-process GPU usage, exposed as nvidia_smi_* Prometheus metrics.
- Supports three collection backends: Native NVML (default on Linux and Windows), nvidia-smi subprocess (default on macOS and Linux -static), and a synthetic demo backend that serves realistic metrics without a GPU.
- The NVML backend exports additional metrics: XID and MIG XID error counters, total energy consumed, per-MIG-instance metrics, GPM activity ratios, PCIe throughput, NVLink and C2C link metrics, per-fan speeds, BAR1 memory usage, thermal thresholds, and power source state.
- Can attribute GPU memory and engine utilization per process to containers (Docker, containerd, CRI-O, Podman), systemd services, or MIG instances on Linux.
- Provides two official Grafana dashboards: Nvidia GPU Metrics (ID 14574) and Nvidia GPU Overview (ID 25547).
- Release artifacts, container images, and Helm charts are signed and carry GitHub build provenance attestations.
Use cases
- Consumer and prosumer GPUs (GeForce/RTX), workstations, and gaming rigs where datacenter fleet tooling is not available.
- Small Kubernetes clusters, edge boxes, and homelabs that need GPU metrics, alerts, and dashboards without installing the NVIDIA GPU Operator stack.
- Virtualized or restricted setups such as vGPU guests, MIG slices, locked-down containers, and WSL2.
- Mixed fleets of old and new cards across Linux and Windows that require a single exporter with a uniform metric schema.
- Remote monitoring of another host over an ssh wrapper command using the exec backend.
Getting started
On a Linux machine with the NVIDIA driver and the NVIDIA Container Toolkit:bashdocker run -d --name nvidia_gpu_exporter --restart unless-stopped --gpus all -e NVIDIA_DRIVER_CAPABILITIES=utility -p 9835:9835 utkuozdemir/nvidia_gpu_exporter:latestcurl http://localhost:9835/metricsPass --collect.backend=exec after the image name to use the nvidia-smi subprocess backend. For Windows, macOS, packages, Kubernetes, and running without Docker, see the repository INSTALL.md.
License and pricing
This project is licensed under the MIT License. It is open source and free to use.
Based on the project's public README and repository metadata.