nvitop CLI
The nvitop terminal monitor shows NVIDIA GPU load, memory and processes to help identify workloads occupying a device.
- 7,240 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The monitor reads process information and can send signals through interactive actions.
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/nvitopDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The monitor reads process information and can send signals through interactive actions.
This is third-party code. Review the repository files before installing.
What it does
nvitop displays NVIDIA GPU status and related processes in a live terminal interface. Users inspect memory and utilization, sort processes and view parent trees to identify a workload. It reads devices through NVML and host processes through psutil. Snapshot output and APIs complement the interactive monitor. The package also includes nvisel for selecting a CUDA device for a new job.
Who it is for. Machine-learning researchers sharing NVIDIA GPUs and inspecting per-process memory use.
Good fit when
- Training runs out of memory and competing GPU jobs need identification.
- Multiple devices have uneven utilization that needs comparison.
- An unfamiliar GPU process needs its parent chain inspected.
- A custom monitor needs device metrics through a Python API.
Not a fit when
- The machine lacks an NVIDIA GPU and compatible NVML.
- You expect automatic model optimization from utilization readings.
- Another process’s environment must not be exposed or the process stopped without approval.
Example request
Inspect which of my processes occupy NVIDIA GPU memory. Show metrics without environment values and terminate nothing.Limitations
Python 3.8 or newer and NVML from a compatible NVIDIA driver are required; the monitor does not prepare GPU drivers. Metric availability depends on the GPU, driver and user permissions, so check the readings you need on your machine. A process environment view can expose secrets; inspect it before including anything in a report. Sending a signal can interrupt computation: identify the process owner and purpose, then agree on the action.
How to disable. Exit the monitor and optionally uninstall nvitop from its Python environment. Observed jobs continue unless you sent them signals.
FAQ
How do I run without a permanent install?
Run `uvx nvitop`. uv prepares a separate Python environment and opens the monitor.
Does nvitop install the NVIDIA driver?
No. NVML must already be available. The repository’s separate driver script is not the normal monitor setup.
Is Windows supported?
Yes. On Windows, the interface uses windows-curses; mouse and display behavior depend on the terminal emulator. Test them on your machine.
Can I inspect GPU-process parents?
Yes. Use the process-tree screen to connect a workload to its originating task.
What is nvisel for?
It helps select a CUDA device. Review its selection conditions before launching a job.
Discussion
No questions or reviews yet.
Related
CLI toolsbottom displays processes, CPU load, memory, network and disks in the terminal with graphs, search and configurable widgets.
A GPU profiler for AI agents: the agent runs vLLM or SGLang under graphsignal-run, reads the metrics and tunes flags for speed
Searchable processes and configurable columns help locate CPU load, listening ports and child commands from a terminal.
witr identifies the process holding a port or file and shows its startup chain, supervising service or container context.
PM2 is a process manager that saves background app lists for restart after reboot and displays their status in the terminal.
A single-container web dashboard for monitoring and controlling a fleet of 1C MCP servers: status, indexing progress, VRAM and license checks
README in short
The monitor provides process filters and historical metric graphs. Dedicated API documentation supports programmatic use.