Load a model, then run ollama ps. Its PROCESSOR column says where the model was loaded: 100% GPU means entirely into the GPU, 100% CPU means entirely into system memory, and a split such as 48%/52% CPU/GPU means partly into both. On a Mac, Ollama's documentation says GPU acceleration uses the Metal API on Apple devices, that its macOS page lists Apple M series chips with GPU support and x86 as CPU only, and that GPU acceleration is not available for Docker Desktop on macOS, so Ollama running in Docker Desktop on a Mac cannot use the GPU. A model that is partly on the CPU is a memory question: the one case this site has read, from an Ollama pull request, is on the page about the Mac's GPU memory limit. This page measures no speed.
Start a model, then in another Terminal window run:
ollama ps
Ollama's FAQ shows output in this form. The example is the FAQ's own: a 42GB model loaded entirely into the GPU.
NAME ID SIZE PROCESSOR UNTIL
llama3:70b bcfb190ca3a7 42 GB 100% GPU 4 minutes from now
| PROCESSOR says | What Ollama's FAQ says it means |
|---|---|
| 100% GPU | The model was loaded entirely into the GPU |
| 100% CPU | The model was loaded entirely in system memory |
| 48%/52% CPU/GPU | The model was loaded partially onto both the GPU and into system memory |
A chip that is not an M series, such as the A18 Pro in the MacBook Neo, is not named on Ollama's list either way; the MacBook Neo page says what has and has not been tested.
Ollama's troubleshooting page gives this command for the server log on a Mac, and the macOS page says the logs are in ~/.ollama/logs.
cat ~/.ollama/logs/server.log
| Claim | Source |
|---|---|
ollama ps, the PROCESSOR column and what each value means, and no GPU acceleration in Docker Desktop on macOS | Ollama FAQ |
| Apple M series with CPU and GPU support, x86 with CPU only, and the log folder | Ollama documentation, macOS |
| GPU acceleration on Apple devices uses the Metal API | Ollama documentation, Hardware support |
cat ~/.ollama/logs/server.log on a Mac | Ollama documentation, Troubleshooting |
ollama ps while a model is loaded and read the PROCESSOR column: 100% GPU is entirely the GPU, 100% CPU is entirely system memory, and a split means partly both.