Skip to content

Can you run Ollama on a 16 GB Mac?

By Junaid, maker of Taskglass · Updated

Summarize this guide with

Yes. On a 16 GB Apple Silicon Mac, 7–8B models at 4-bit quantization (about 5 GB) run well alongside normal apps, 12–14B models (about 8–9 GB) fit if you close heavy apps, and 30B and larger models don't fit. Ollama keeps a model in memory for a few minutes after use; ollama stop frees it right away.

How to run Ollama well on 16 GB

  1. 1

    Pick a model that fits

    Choose a 7–8B model for everyday use. The model's download size is a good guide to the memory it needs, plus some extra for the conversation (context).

  2. 2

    Check where it runs

    While the model is loaded, ollama ps shows its size and whether it runs fully on the GPU. If part of it falls back to the CPU, it's too big for comfort: use a smaller model.

  3. 3

    Keep the context reasonable

    Longer context windows use more memory on top of the model. Very large context settings can push a model that fit into swapping.

  4. 4

    Free the memory when you're done

    ollama stop <model> unloads it immediately. By default Ollama unloads a model after about 5 minutes idle; the OLLAMA_KEEP_ALIVE setting changes that.

Model sizes on a 16 GB Mac (4-bit)

Model sizeMemoryOn 16 GB
1–4B1–3 GBRuns easily
7–8Babout 5 GBRuns well with your apps open
12–14Babout 8–9 GBFits if you close heavy apps
30–32Babout 18–20 GBDoesn't fit
70Babout 40 GBDoesn't fit

Why is Ollama slow on my Mac?

Usually the model is too big for your memory, so part of it runs on the CPU or macOS swaps. ollama ps shows the CPU/GPU split; a smaller model or closing heavy apps fixes it.

Does Ollama use the GPU on Apple Silicon?

Yes, through Metal. Apple Silicon shares one pool of memory between CPU and GPU, and by default macOS lets the GPU use only part of it, which is why a model close to your total RAM won't fit.

Can I run Ollama on an 8 GB Mac?

Small models (1–4B) run; 7–8B models are tight and will swap with other apps open.

How Taskglass helps

Taskglass's Agents tab shows Ollama and LM Studio with the memory their loaded model holds, whether they're busy or idle, and flags a model sitting idle in memory in its diagnosis.

See how it works

Questions

How much RAM does Llama 3.1 8B need?

About 5 GB at the default 4-bit quantization, plus a little more for context.

How do I unload a model in Ollama?

Run ollama stop <model>. ollama ps lists the models currently loaded.

Is 16 GB enough for local AI and coding at the same time?

With a 7–8B model, yes, for most setups. Add Docker or simulators and 24–32 GB is more comfortable.

Processes you might see

More guides

See the cause on your own Mac.

Taskglass shows which app is slowing, heating or draining your Mac, and explains it in plain English. One-time purchase, 30-day money-back guarantee.