OllamaThe GPU is not being used

Ollama shows 100% CPU instead of the GPU

What you will see

100% CPU

What it means

ollama ps shows where a model is loaded. "100% CPU" means none of it is on the GPU; a split such as "48%/52% CPU/GPU" means it did not fit.

Why it happens

  • Ollama found no usable GPU: driver problem, container without GPU access, or an AMD card outside ROCm's list.
  • A split figure: the model and its context are larger than the card.

How to fix it

  1. Read the server log for the GPU discovery lines (Ollama's troubleshooting doc lists where the log lives on each OS).
  2. For a split, check the size in the calculator and pick a smaller level, or lower num_ctx.
  3. AMD on Linux: an unlisted Radeon may run with HSA_OVERRIDE_GFX_VERSION (see that page).

Check the size before changing flags

Most memory errors are a size problem. Put the model, quant level, context and your card into the VRAM calculator: if it is red there, no flag will make it fit.

Open the VRAM calculatorLongest context on common cards

Guides that cover this

Related errors

Checked against

Message text read from source on 2026-10-26.