Repolode
Submit
psalias2006

psalias2006/gpu-hot

Self-hosted, web-based real-time dashboard for monitoring NVIDIA GPUs, with a hub mode for multiple machines.

View on GitHubWebsite
Stars
1,640
Forks
85
Open issues
8
Last commit
2 months ago

What is it?

GPU Hot is a web dashboard for watching NVIDIA GPU metrics in real time. It runs as a Docker container, reads GPU data through NVML with an nvidia-smi fallback for older cards, and serves the dashboard on port 1312 so you can view it from any browser.

The dashboard shows per-GPU utilization, temperature, memory, power draw, fan speed, clock speeds, PCIe info, P-State, throttle status, and encoder/decoder sessions, along with GPU process names and PIDs and host CPU and RAM. Historical charts cover utilization, temperature, power, and clocks.

For clusters, a hub instance with no GPU can aggregate several GPU nodes listed in NODE_URLS. It also exposes JSON and WebSocket endpoints so the metrics can be consumed by other tools.

Key features

  • 01Real-time GPU metrics with sub-second updates
  • 02Automatic multi-GPU detection
  • 03Process monitoring showing PID and memory usage
  • 04Historical charts for utilization, temperature, power and clocks
  • 05Hub mode to aggregate multiple GPU nodes

Use cases

  • Watching utilization and temperature on a single local GPU workstation
  • Monitoring a small cluster of GPU servers from one hub dashboard
  • Checking which processes are using GPU memory during training or inference

Quick start

  1. 1

    Make sure Docker and the NVIDIA Container Toolkit are installed

  2. 2

    Run GPU Hot on a single machine with Docker

    docker run -d --gpus all -p 1312:1312 ghcr.io/psalias2006/gpu-hot:latest
  3. 3

    Open the dashboard in your browser

Things to know

  • Requires NVIDIA GPUs and the NVIDIA Container Toolkit, so it does not work on other GPU vendors.
  • Process names need the container to run with host PID access, which exposes host process information.

Topics

#gpu-monitoring#nvidia#dashboard#docker#real-time-monitoring#system-monitoring#charts#cuda#devops#flask

FAQ

What is gpu-hot?
Self-hosted, web-based real-time dashboard for monitoring NVIDIA GPUs, with a hub mode for multiple machines. GPU Hot is a web dashboard for watching NVIDIA GPU metrics in real time. It runs as a Docker container, reads GPU data through NVML with an nvidia-smi fallback for older cards, and serves the dashboard on port 1312 so you can view it from any browser.
Is gpu-hot free to use? What license does it have?
gpu-hot is open source under the MIT license. Check the license terms before using it in commercial products.
Is gpu-hot still maintained?
Updated within 6 months. The last commit to gpu-hot was on Aug 25, 2026.
What can I use gpu-hot for?
Watching utilization and temperature on a single local GPU workstation Monitoring a small cluster of GPU servers from one hub dashboard Checking which processes are using GPU memory during training or inference
How do I install gpu-hot?
Make sure Docker and the NVIDIA Container Toolkit are installed Run GPU Hot on a single machine with Docker: docker run -d --gpus all -p 1312:1312 ghcr.io/psalias2006/gpu-hot:latest Open the dashboard in your browser
Where can I get gpu-hot?
The source code and installation instructions are on GitHub: https://github.com/psalias2006/gpu-hot

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects