Repolode
Submit
leptonai

leptonai/gpud

A Go daemon that continuously monitors NVIDIA GPUs and GPU fabric, detects hardware errors, and reports node health on Linux hosts and Kubernetes.

View on GitHubWebsite
Stars
498
Forks
66
Open issues
4
Last commit
yesterday

What is it?

GPUd is a self-contained binary that watches GPUs on Linux machines and surfaces their health, metrics and faults in one place. It collects power and temperature readings, parses nvidia-smi output for GPU and fabric status, and detects errors from kernel messages, hardware slowdown events, NVML Xid events and DCGM.

Beyond GPUs, it also tracks overall system metrics such as CPU, memory and disk usage. Monitoring is designed to run off the critical path of AI/ML workloads, keeping CPU and memory overhead low.

GPUd can run as a systemd service on a host, in the foreground, or be deployed to Kubernetes with an official Helm chart. It is also used in the production infrastructure of NVIDIA DGX Cloud Lepton, which can register machines with it.

Key features

  • 01Collects GPU power and temperature metrics and GPU fabric metrics.
  • 02Reports GPU and fabric status by parsing nvidia-smi output.
  • 03Detects errors from kernel logs, NVML Xid events, hardware slowdown and DCGM.
  • 04Tracks CPU, memory and disk usage on the host.
  • 05Ships as a single binary with systemd, foreground and Kubernetes Helm deployment options.

Use cases

  • Monitoring health and detecting faulty GPUs across a fleet of Linux GPU servers.
  • Running GPU health checks on Kubernetes clusters via the Helm chart.
  • Diagnosing GPU errors such as Xid faults or hardware slowdown on AI training nodes.
  • Registering GPU machines with NVIDIA DGX Cloud Lepton.

Quick start

  1. 1

    Run the official install script from the README (Linux amd64 or arm64)

  2. 2

    Start the GPUd service with systemd

    sudo gpud up

Things to know

  • Targets Linux hosts with NVIDIA GPUs; it is not for macOS or Windows.
  • Commands need root privileges (sudo).
  • Host installs enable automatic updates by default.

In collections

Run and train AI on GPUsOpenTelemetry & observability platforms

Topics

#gpu#nvidia#monitoring#kubernetes#diagnostics#observability#linux#nvidia-gpu

FAQ

What is gpud?
A Go daemon that continuously monitors NVIDIA GPUs and GPU fabric, detects hardware errors, and reports node health on Linux hosts and Kubernetes. GPUd is a self-contained binary that watches GPUs on Linux machines and surfaces their health, metrics and faults in one place. It collects power and temperature readings, parses nvidia-smi output for GPU and fabric status, and detects errors from kernel messages, hardware slowdown events, NVML Xid events and DCGM.
Is gpud free to use? What license does it have?
gpud is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is gpud still maintained?
Actively maintained. The last commit to gpud was on Nov 5, 2026.
What can I use gpud for?
Monitoring health and detecting faulty GPUs across a fleet of Linux GPU servers. Running GPU health checks on Kubernetes clusters via the Helm chart. Diagnosing GPU errors such as Xid faults or hardware slowdown on AI training nodes. Registering GPU machines with NVIDIA DGX Cloud Lepton.
How do I install gpud?
Run the official install script from the README (Linux amd64 or arm64) Start the GPUd service with systemd: sudo gpud up
Where can I get gpud?
The source code and installation instructions are on GitHub: https://github.com/leptonai/gpud

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

Related projects