Aug 25, 2025 20:30:00

'nvitop' is a tool that shows NVIDIA GPU processes in real time in a more understandable way than nvidia-smi

Developers and researchers using NVIDIA GPUs often use the '

nvidia-smi ' command, NVIDIA's system management interface. However, sometimes you want to manage processes more interactively or grasp information graphically. ' nvitop ' features a colorful, easy-to-understand interface that displays information not just once, but also continuously updates information as a resource monitor.

GitHub - XuehaiPan/nvitop: An interactive NVIDIA-GPU process viewer and beyond, the one-stop solution for GPU process management.
https://github.com/XuehaiPan/nvitop

Welcome to nvitop's documentation! — nvitop: the one-stop solution for GPU process management. documentation
https://nvitop.readthedocs.io/en/latest/

nvitop has similar functionality to nvidia-smi, but it also displays continuously updated information about your GPU's status through a colorful, informative interface. It aims to be a one-stop solution for GPU management, going beyond simple monitoring to include process filtering, sorting, a tree view, and even providing an API that can be used by other programs.

nvitop displays more information than nvidia-smi in an easy-to-read, color-coded box format. The colors change depending on GPU and memory usage, allowing you to intuitively understand the system load.

The most notable feature of nvitop is its monitor mode, which continuously updates information. In this mode, it responds to keyboard and mouse input, and allows you to change the sort order of processes and filter only processes belonging to specific users. You can also switch to a process tree view, which allows you to safely terminate processes while checking their parent-child relationships.

Select a specific process and press Enter to display various metrics, such as CPU usage and GPU memory usage, of that process in a time series graph, allowing you to analyze the resource consumption trends of the process in detail.

nvitop runs on both Linux and Windows, allowing you to monitor your GPU with consistent operability across different OS environments.

nvitop also comes with a command line tool called 'nvisel' that automatically selects available GPUs based on conditions such as memory size and utilization. It also provides an API that allows developers to incorporate GPU monitoring functionality into their own applications, making it highly extensible.

nvitop is written in Python and is recommended to be run in a virtual environment using uv or pipx. The installation command using uv is as follows:
[code]uvx nvitop[/code]

If you are using pipx, you can install it with the following command:
[code]pipx run nvitop[/code]

You can also install it with pip.
[code]pip3 install --upgrade nvitop[/code]

Additionally, nvitop requires Python 3.8 or later and the NVIDIA display driver or the NVIDIA Management Library (NVML) included in the CUDA Toolkit.