prometheus_exporter_gpu

Simple Prometheus exporter that collects and exposes GPU metrics from the local host for scraping by Prometheus.

Features

• Exposes GPU utilization, memory, temperature and driver information.
• HTTP metrics endpoint compatible with Prometheus (/metrics).
• Small, single-binary or container-friendly.

Prerequisites

• Docker

Quick start

docker compose up -d

Default metrics endpoint: http://localhost:8000/metrics

Metrics

Exposed metrics:

• power_w
• gpu_temp_c
• gpu_clock_mhz
• mem_clock_mhz
• fan_speed_percent
• gpu_util_percent
• mem_util_percent
• memory_used_mib
• memory_total_mib

Check /metrics for the full list from the running exporter.

Prometheus scrape config (example)

- job_name: 'gpu_exporter'
    static_configs:
        - targets: ['host.example:8000']