rustcost-gpu-exporter

A lightweight, high-performance GPU metrics exporter for NVIDIA & AMD GPUs.
Deployable as a standalone container or as a Kubernetes DaemonSet using the official Helm chart.

---

Overview

rustcost-gpu-exporter exposes real-time GPU metrics over HTTP in JSON.

It is designed to be:

• Extremely fast (Rust)
• Tiny footprint (milliseconds CPU time)
• Supports NVIDIA & AMD GPUs
• Deployable on Kubernetes or standalone containers

---

Usage Options

You can use rustcost-gpu-exporter in two ways:

---

1. Deploy on Kubernetes

(DaemonSet via Helm Chart)
Recommended: Use with rustcost Core + Dashboard

For a complete cost monitoring experience
including dashboards, alerting, and extended rustcost features
install the rustcost core Helm chart, which includes ready-to-use dashboards:

https://github.com/rustcost/rustcost-helmchart

---

2. Standalone Docker Container

Prerequisites for NVIDIA GPUs on Linux

When using NVIDIA GPUs on Linux systems, the following must be installed:

• NVIDIA Driver: NVIDIA GPU driver must be installed on the host system
• NVIDIA Container Toolkit: Required to expose GPUs to Docker containers

Running the Container

Run manually on any Linux system:

docker run -d \
  --gpus all \
  -p 8000:8000 \
  -e GPU_EXPORTER_COMPLEX=0 \
  -e PORT=8000 \
  -e COLLECT_INTERVAL_SEC=60 \
  rustcost/gpu-exporter:0.0.2

Metrics now available at:

http://localhost:8000/metrics

---

Environment Variables

Variable	Default	Description
GPU_EXPORTER_COMPLEX	0	Enable heavy/complex metrics
PORT	8000	Web server port
COLLECT_INTERVAL_SEC	60	Scrape refresh interval

---

Example Output

{
  "amd": [],
  "nvidia": [
    {
      "gpu_memory_total_mb": 4096,
      "gpu_memory_used_mb": 565,
      "gpu_utilization_percent": 15,
      "index": 0,
      "name": "NVIDIA GeForce RTX 3050 Laptop GPU",
      "up": true
    }
  ]
}

---

Contributing

We welcome:

✔ Issues ✔ Feature requests ✔ Pull requests

---

License

Licensed under the Apache License, Version 2.0.