Local AI GuideGPU support in Ollama
Ollama uses your GPU if it supports it. The documentation lists what it supports.
Read on 14 November 2026. This field changes fast: check the sources.
What the page says
| Hardware | Support, as the page states it |
|---|---|
| Nvidia | GPUs with compute capability 5.0 or higher and driver version 550 or newer. Cards with compute capability 5.0 to 6.2 need driver 570 or newer. The page links Nvidia's list to check your card |
| AMD Radeon | Via ROCm. On Linux it needs the AMD ROCm v7 driver. The page lists supported Radeon RX, PRO and accelerator families |
| Apple | GPU acceleration through the Metal API |
| Other GPUs | More GPU support on Windows and Linux through Vulkan, which is enabled by default when the backend is installed |
Notes from the page
- On Windows, most GPU vendors' drivers come with Vulkan support. On Linux you may need extra components.
- The scheduler uses the VRAM data that the GPU libraries report. Vulkan needs extra capabilities or root to report it; without them Ollama uses approximate model sizes for scheduling.
- To choose Vulkan GPUs, set
GGML_VK_VISIBLE_DEVICES. To switch Vulkan off, setOLLAMA_VULKAN=0.
Checking your own computer
- Find your card's model in your system information.
- Look it up in Nvidia's compute capability list or in the page's AMD list.
- Update the driver to the version named.
- Start a model and run
ollama ps: the PROCESSOR column shows how much runs on the GPU.
My card is not listed.
It may still work through Vulkan. Check the page and test with ollama ps.
Does it work without a GPU?
Yes: the page speaks of falling back to the CPU, which is slower.
Sources
- Ollama documentation: Hardware support (docs.ollama.com/gpu)
Read on 14 November 2026.
This is an independent guide. It is not connected with Ollama, the llama.cpp project or any company named here. This field changes quickly: statements about Ollama and llama.cpp were read on 14 November 2026 from the pages named under Sources. The memory numbers are estimates from arithmetic. We did not run any model for these pages.