picomon 0.1.1

Minimal curses dashboard for monitoring AMD GPUs via amd-smi

picomon

picomon is a tiny terminal dashboard for monitoring AMD GPUs via amd-smi. It polls basic metrics (gfx activity, memory usage, and power) and renders them as sparklines inside a curses UI so you can keep an eye on accelerators without launching a full GUI stack.

Homepage: https://omarkamali.github.io/picomon/

Why?

I like nvtop but the asserts kept crashing it on some AMD devices. picomon is a lightweight alternative that just polls metrics and renders them as sparklines, trading off ironclad accuracy checks for more reliability.

It hasn't been tested on all AMD GPUs. If it fails to run on your GPU, please open a new issue using this template.

Requirements

• Python 3.9 or newer
• The amd-smi CLI available on your PATH (if your GPU is properly setup then you already have this)
• An ANSI-compatible terminal for the curses UI

Installation

pip install picomon

Usage

After installation, run the CLI:

$ picomon

┌──────────────────────────────────────────┐  ┌──────────────────────────────────────────┐
│ GPU 0  GFX  42%  UMC  21%                │  │ GPU 1  GFX  78%  UMC  66%                │
│ PWR 135/250W (54%)  VRAM 10.0/16.0GB 62% │  │ PWR 210/250W (84%)  VRAM 14.5/16.0GB 90% │
│                                          │  │                                          │
│ GFX ▁▂▂▃▄▄▅▆▆▇█▇▆▅▄▃▂▁                   │  │ GFX ▂▃▄▅▆▇██▇▆▅▄▂▂▃▅▆                    │
│ PWR ▁▁▂▂▃▄▄▅▆▇██▇▆▅▄▂▁                   │  │ PWR ▂▂▃▄▅▆▇██▇▆▅▄▃▂▂▃                    │
│ VRM ▁▁▂▂▃▄▄▅▆▇███▇▆▅▄▂                   │  │ VRM ▂▃▄▅▆▆▇███▇▆▅▄▃▂▂▃                   │
└──────────────────────────────────────────┘  └──────────────────────────────────────────┘

Key bindings:

• q to quit

Common flags:

• --update-interval (seconds between refreshes, default 3)
• --history-minutes (rolling window to retain, default 30)
• --static-timeout / --metric-timeout (seconds to wait for amd-smi responses)

These correspond to the PicomonConfig dataclass, so you can also import and reuse picomon as a library:

from picomon import PicomonConfig, run_monitor

config = PicomonConfig(update_interval=1.5)
run_monitor(["--update-interval", str(config.update_interval)])

Development

• Run tests with pytest (see the CI workflow for reference)
• Use scripts/publish.py to cut a GitHub release once tags are in place

License

MIT © Omar Kamali

Source: https://github.com/omarkamali/picomon