NVIDIA/dcgm-exporter

NVIDIA GPU metrics exporter for Prometheus leveraging DCGM

NVIDIA DCGM Exporter – What it is

The DCGM Exporter is a small service that reads telemetry from NVIDIA’s Data Center GPU Manager (DCGM) and makes that data available in the Prometheus text format. In practice, it lets you scrape GPU health, utilization, temperature, power draw, memory usage and many other counters with any Prometheus‑compatible monitoring stack (Grafana, OpenObserve, etc.).

Core purpose

  • Expose GPU metrics: Pulls a wide set of counters from DCGM and presents them as Prometheus metrics.
  • Integrate with observability tools: Comes with Helm charts, container images, and ready‑made Grafana dashboards for quick deployment in Kubernetes or on bare‑metal servers.

Main components shipped in the repo

Component What it contains
Exporter source Go code that talks to the DCGM library and formats metrics.
Container & package builds Docker images and package manifests for easy installation.
Helm chart Kubernetes manifests and configuration helpers (deployment/README.md).
Tests End‑to‑end Kubernetes tests (tests/k8s/README.md) and container‑level tests (tests/container/README.md).
Dashboards Grafana dashboard JSON (grafana/dcgm-exporter-dashboard.json) and links to community dashboards (Grafana.com, OpenObserve).
Default metric list CSV (etc/default-counters.csv) that defines which counters are collected out of the box.

How to get started (as described in the README)

  1. Install – Follow NVIDIA’s official install guide linked in the README.
  2. Configure Prometheus – Add the exporter endpoint to your Prometheus scrape config (link provided).
  3. Verify – Use the command‑line reference to check that metrics are being emitted.
  4. Visualise – Import the supplied Grafana dashboard or use the community OpenObserve dashboard.

Compatibility notes

  • The exporter must be paired with the same DCGM version it was built for. The README points to a version‑compatibility page that lists supported DCGM‑exporter combinations and remote‑connection options.

Contributing & support

  • Contribution guidelines are in CONTRIBUTING.md.
  • Issues and community help are handled via the GitHub issues page.
  • Security reports should go through NVIDIA’s Product Security process (see SECURITY.md).

Who would use this?

  • System administrators managing GPU‑heavy clusters (AI training, inference, HPC).
  • DevOps/Observability engineers who need Prometheus‑compatible metrics for alerting and dashboards.
  • Developers building custom monitoring or auto‑scaling logic around GPU health.

Bottom line: The DCGM Exporter is a production‑ready bridge between NVIDIA’s GPU management layer and the Prometheus ecosystem, enabling reliable monitoring of GPU resources in AI and high‑performance computing environments.

Related

  • Project
    utkuozdemir/nvidia_gpu_exporter

    A Prometheus exporter that collects NVIDIA GPU metrics using nvidia-smi or NVML, supporting Linux, Windows, and macOS.

  • Project
    NVIDIA/DCGM

    A suite of tools for managing and monitoring NVIDIA data-center GPUs in cluster environments, providing health monitoring, diagnostics, and governance policies.

  • Project
    wentbackward/nv-monitor

    A lightweight, zero-dependency monitoring tool for NVIDIA GPU systems that provides a TUI, CSV logging, and a Prometheus exporter for tracking CPU, GPU, and memory metrics.

  • Project
    leptonai/gpud

    A production-grade GPU monitoring and management tool that ensures hardware efficiency and reliability for AI/ML workloads.

  • Project
    NVIDIA/k8s-device-plugin

    The NVIDIA k8s‑device‑plugin is an official DaemonSet that implements Kubernetes’ device‑plugin API for NVIDIA GPUs. It discovers GPUs, reports health, and makes them schedulable via the `nvidia.com/gpu` resource (with optional MIG and shared‑GPU modes). Install the NVIDIA driver and container toolkit on each node, deploy the plugin (static manifest or Helm), then request GPUs in pod specs. Configuration is possible via flags, env‑vars, or a YAML file, covering MIG strategy, device‑list delivery, shared‑GPU (time‑slicing or MPS), and more. The plugin is beta, lacks full health‑check/cleanup, and only the official NVIDIA version is supported.