htop-gpu 0.1.0

Terminal dashboard for NVIDIA GPUs, system CPU/memory, and processes — clickable, with conda/docker/tmux info per process.

htop-gpu

NVIDIA GPUs, system CPU/memory, and processes in one terminal dashboard — each process annotated with its conda env, docker container, and working directory.

Install · Usage · Keys · Why

Install

pip install htop-gpu

htop-gpu          # one-shot snapshot
htop-gpu -w       # watch mode
hgpu -w           # short alias

Linux, Python 3.9+, NVIDIA driver. Uses NVML (fast) when available, falls back to nvidia-smi.

What it does

• GPUs, CPU/memory, and processes in one screen
• Click any panel title to switch what the process table shows (cpu / memory for top system processes, gpus for GPU-only, processes to fullscreen the table)
• Click column headers to sort
• Click a process row → press k to send SIGTERM (falls back to sudo kill with a password prompt if you don't own it)
• Conda env, docker container, tmux session shown next to each command, so you can tell at a glance whose stray process is hogging GPU 5
• Adaptive layout — drops decorations as the terminal shrinks; it keeps the GPU + CPU/Mem panels visible even on tiny windows

Usage

Watch mode is mouse-driven. Everything is also keyboard:

key	what it does
c / m / p	switch to cpu / memory / focus-procs view
Esc, ←	back out (clears mode / focus / selection)
↑ ↓	move selection
k, F9	kill selected process
l	toggle full command lines
0–9	filter to that GPU index
q, F10, Ctrl-C	quit

Mouse uses SGR mouse mode (\x1b[?1006h) — works in iTerm2, Kitty, WezTerm, Ghostty, Alacritty, Windows Terminal, modern xterm/Konsole, and tmux with mouse on.

For scripts:

htop-gpu --json | jq '.processes[] | select(.gpu_mem_mib > 10000)'

Credits

UI takes inspiration from htop, btop, and nvtop. Independent project, not affiliated.

License

MIT