Ollama Hardware Requirements

Ollama hardware requirements depend almost entirely on the size of the model you want to run. A model has to fit in memory: ideally in your GPU VRAM, otherwise in system RAM where it runs more slowly.

Model size is measured in parameters, written as 1B, 3B, 8B, 70B and so on (B = billion). Ollama library models are usually 4-bit quantized by default, which shrinks memory use a lot.

Rough Memory Guide

  • 1B - 4B models: about 1-3 GB. Run on almost any laptop, even CPU only.
  • 7B - 9B models: about 5-6 GB. Comfortable with 8 GB VRAM or 16 GB RAM.
  • 12B - 14B models: about 8-10 GB. Best with a 12-16 GB GPU.
  • 27B - 32B models: about 18-22 GB. Needs a 24 GB GPU or a 32 GB+ Apple Silicon Mac.
  • 70B models: about 40+ GB. High-end GPUs or 64 GB+ unified memory.

These figures are approximate for 4-bit quantized models at the time of writing (2026). Longer context windows need extra memory on top, because the model keeps a cache of the conversation.

CPU vs GPU

Ollama works on CPU only, but token generation is much slower: a small model may produce a few words per second instead of dozens. A GPU with enough VRAM is the single biggest speed boost. Apple Silicon Macs are a special case: the GPU shares unified memory with the system, so a Mac with 32 GB can run fairly large models.

Check What Is Loaded

The ps command shows loaded models, how much memory they use, and whether they run on GPU, CPU or a split of both:

ollama ps
NAME              ID              SIZE      PROCESSOR    UNTIL
llama3.2:latest   a80c4f17acd5    4.0 GB    100% GPU     4 minutes from now

If PROCESSOR shows something like 48%/52% CPU/GPU, the model did not fully fit in VRAM and part of it runs on the CPU. Pick a smaller model or a smaller quantization to get it back to 100% GPU.

Choosing a Model Size

  • Start small (1B-4B) to learn the tools and test your pipeline.
  • Use 7B-14B models for everyday chat, coding help and summarisation.
  • Go bigger only when quality clearly matters and your hardware allows it.
  • Prefer a smaller model at full GPU speed over a big one crawling on CPU.

Frequently asked questions

Can I run Ollama without a GPU?

Yes. Ollama runs on the CPU when no supported GPU is found. Small models are usable; large models will be slow.

How much RAM do I need for an 8B model?

Around 8 GB of free memory is a comfortable minimum for a 4-bit 8B model, and 16 GB of system RAM is recommended.

Does Ollama support AMD GPUs?

Yes, Ollama supports many AMD GPUs through ROCm on Linux and Windows, as well as NVIDIA GPUs via CUDA and Apple Silicon via Metal.