No llama-server, 9.5 GB free | Generated by AI
No, llama-server is not running. Nothing matching llama at all. The only AI daemon is ollama serve (PID 3339), and it’s idle — not loaded on the GPU.
Current GPU state (RTX 4070, 12 GB):
| MiB | |
|---|---|
| Used | 2367 |
| Free | 9473 |
And the 2.3 GB used is not AI — it’s all desktop/graphics:
- Xorg: 1430 MiB (biggest)
- warp-terminal: 204 MiB
- zed-editor: 191 MiB
- browsers / gnome-shell / nautilus etc.: rest
So there’s nothing to kill for your use case — you actually have ~9.5 GB free, which is plenty. Your large-v3 fp16 transcription needs only ~3.2 GB (or ~1.1 GB with --compute-type int8). The --low-memory rewrite fits easily.
If you still want headroom, the only meaningful targets would be closing browser/zed/warp-terminal tabs (they’re the desktop apps eating VRAM), but it’s not necessary for the whisper run.
Generated by AI. Curating and sharing still takes effort.
If you find it useful, feel free to
donate.
WeChat: @lzwjava
·
X: @lzwjava ·
Say hi 👋