nvtop
GPU & Accelerator process monitoring for AMD, Apple, Huawei, Intel, NVIDIA and Qualcomm
NVTOP is a terminal dashboard for monitoring GPU usage, memory, temperature, and power draw in real time, similar to htop for CPUs, supporting NVIDIA, AMD, Intel, Apple, and other GPU vendors.
NVTOP is a terminal-based monitoring tool for GPUs and AI accelerators, similar in spirit to the htop utility that Linux users use to watch CPU and memory usage. Instead of showing CPU processes, it shows what is running on your graphics card: how much of the GPU is being used, how much memory it is consuming, temperature, power draw, and which programs are using it.
The tool supports a wide range of hardware from multiple vendors, including AMD, Apple (M1 and M2 with limited support), Huawei Ascend, Intel, NVIDIA, Qualcomm Adreno, Broadcom VideoCore, Rockchip, and several specialized AI accelerator brands. It can display information for multiple GPUs at the same time.
NVTOP runs in a terminal window and displays a live-updating dashboard with per-process GPU usage. It has a built-in setup window accessible by pressing F2, where you can configure what information is shown and how the layout looks. Preferences can be saved so they load automatically the next time you run the program. A manual page is included for reference.
Installing it on Linux is straightforward: most major distributions package it directly, so a single apt, dnf, or pacman command is enough. Snap, AppImage, and conda-forge options are also available. For systems without a package manager or for building from source, the README includes instructions using CMake. Docker and WSL2 setups are covered as well.
Each GPU vendor section in the README notes which kernel version or driver library is required for process-level monitoring to work, since some features depend on interfaces that were added in specific Linux kernel releases. The project is written in C and released under an open-source license.
Where it fits
- Monitor GPU memory and utilization in real time while training AI models to catch memory leaks or underutilization.
- See which processes are consuming GPU resources on a shared server to diagnose slowdowns or conflicts.
- Watch power draw and temperature across multiple GPUs simultaneously during a long compute job.
- Install a lightweight GPU monitor on a headless Linux server without needing a graphical desktop environment.