Open-source GPU observability

Your GPUs.
At a glance.

A fast, keyboard-first terminal monitor for GPU infrastructure. See what is happening now — and rewind to understand what happened before.

Get started → View source ↗

Built forLinux + NVIDIAmacOS + Apple silicon

gputop · gpu-node-01LIVE

1 Overview 2 GPUs 3 Processes 4 Memory

Fleet 8 GPUs

Allocated 8/8

Health 96 avg

Power 3,926 W / 5,600 W

VRAM 447 GiB / 640 GiB

Live 10m

▁▁▂▂▃▃▄▃▅▄▆▅▄▃▄▅▇▆▅▆▇█▇▆▅▅▆▅▄▃▃▄▅▆▇

avg utilization 71%

▂▃▃▄▅▄▅▅▆▅▃▄▃▄▅▇▆▆▅▆▇▇█▇▆▅▅▆▅▄

total power 3926 W

GPUs 8 GPUs

# NAME UTILVRAMTEMPPOWERHEALTH

0 H100 80GB HBM ███████████▊ 98%66.2 / 80G72°C641 W100 ●

1 H100 80GB HBM ███████████▋ 97%66.0 / 80G70°C653 W100 ●

3 ⚠ H100 80GB HBM ███████▏░░░░ 55%66.4 / 80G54°C396 W75 ◐

↑ select ⏎ detail h history p pause ? help q quit

LIVE TELEMETRY◆TIME MACHINE HISTORY◆HEALTH SIGNALS◆KUBERNETES AWARE◆LIVE TELEMETRY◆TIME MACHINE HISTORY◆

Built for the console

Signal, not noise.

Everything you need to understand GPU behavior, arranged for fast scanning and quick investigation.

One clear fleet view

Utilization, VRAM, temperature, power, health and state — per GPU, in a single dense overview.

↶

Rewind with history

Scrub metrics and events through a configurable on-disk time machine. Find the moment a system changed.

⌁

Trace every workload

Follow GPU to process, container, pod and workload. Spot idle allocations and stragglers.

◈

Know what needs attention

Health scores and events surface Xids, throttling, ECC, PCIe and NVLink issues with context.

What gputop tracks

The whole GPU story.

gputop combines official vendor metrics with clearly-labelled derived insights. Unsupported values stay N/A, never misleading zeroes.

Explore all metrics →

01Compute & memoryGPU utilization, VRAM, clocks, processes, MIG

02Power & thermalsPower draw, energy, temperature, limits, throttle reasons

03Fabric & I/ONVLink, PCIe state, throughput, topology and errors

04Health & eventsECC, Xids, row remapping, retired pages and alerts

05InfrastructureCPU, RAM, disks, network, InfiniBand and Kubernetes

Made to investigate

Sixteen focused views, one fast workflow.

gputop overview with fleet metrics, live charts, GPU table and events

Start in seconds

Meet your GPU fleet.

Download

Grab the latest release for Linux or macOS.

Install

Make it executable and put it on your PATH.

Run

Launch the TUI — or try the full demo without hardware.

Terminal

$ curl -fLo gputop https://github.com/gputop/gputop/releases/latest/download/gputop-linux-amd64
$ chmod +x gputop && sudo install gputop /usr/local/bin/
$ gputop --demo

Have Go 1.25+? Run go install github.com/gputop/gputop/cmd/gputop@latest.

One binary, many ways to use it

gputopInteractive GPU terminal

gputop --demoExplore simulated GPUs

gputop --once --jsonOne machine-readable snapshot

gputop --serviceHTTP API + Prometheus agent

gputop --remote gpu-node-01Connect to a remote agent

gputop --print-configView effective configuration

Documentation

Everything, explained.

Guides for running gputop locally, in production, and across your GPU fleet.

ConfigurationSet refresh tiers, retention, themes, keys and server options.→Metrics referenceUnderstand each collected and derived metric.→Health scoreLearn how gputop identifies risk and explains it.→KubernetesDeploy as a DaemonSet and correlate workloads.→Remote monitoringConnect agents securely across nodes.→ThemesMake the terminal look exactly the way you work.→

Open source

Bring clarity to your compute.

Star gputop on GitHub ↗

Your GPUs.At a glance.