GitHub radar
See exactly what's using your GPU's memory
GPUTaskMan shows which processes are using your GPU's memory and how much — instead of one vague number, you get the real picture.
A terminal tool that shows GPU usage broken down by process. For AI models, it separately shows how much memory went to model weights versus buffers.
When it helps
Useful if you run AI models locally and run short on GPU memory. Not needed if your models run in the cloud or you have no memory issues.
Pros
- Free, no account — just reads your GPU's own data
- Works on Windows, Linux, macOS, and inside WSL
- Recognizes models running through Ollama and LM Studio
- Splits a model's memory into weights and buffers
Cons
- Author calls it early-stage, experimental software
- Per-model memory estimates aren't always accurate
- Only diagnoses the problem, doesn't fix it
How to set it up — step by step
- 1Open your AI agent (like Claude Code) in your project's terminal folder.
- 2Send the agent the text from the block below to install and run the tool.
- 3Wait for install: pipx install gputaskman or uv tool install gputaskman.
- 4Run the gputaskman command in your terminal and check the process list.
- 5See which process uses the most GPU memory — model, game, or browser.
- 6If memory's short, close that process and restart your local model.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install gputaskman from https://github.com/shanselman/gputaskman (via pipx install gputaskman or uv tool install gputaskman) and run it, then show me what is currently using up my GPU memory.
The tool itself is lightweight — it just needs a computer with an NVIDIA or AMD GPU (Windows, Linux) or an Apple Silicon Mac; no extra memory is needed, it only reports what's already running on your own machine.
Open on GitHub▌ More finds