Skip to main content

macOS setup (Apple Silicon)

Requirements​

  • Apple Silicon (M1, M2, M3, M4) — Intel Macs are not supported for the text worker
  • 32GB+ unified memory — 16GB and 24GB Macs are below the bar, they can't hold the model
  • Node.js 18+
  • ollama v0.32.15 or newer

Install Node.js​

Using Homebrew:

brew install node

Or download from nodejs.org.

Install (or upgrade) ollama​

The worker checks the ollama version at startup and stops if it's older than 0.32.15 — older builds return unhelpful HTTP 500s instead of loading the model:

brew upgrade ollama    # or: brew install ollama, or ollama.com/download
ollama --version

Run the worker​

npx @compute-network/worker --token <your-token>

That's it. ollama automatically detects Metal on Apple Silicon — no extra drivers or configuration needed.

There is nothing to choose: every native worker serves the same model, qwen3.8-27b-uncensored. On a Mac you get the GGUF noMTP build, run by ollama on Metal (speculative decoding is never used on Metal). The old --model flag is deprecated and ignored; pass --mode max to skip the text/image question.

Get a token at compute.tech/earn.

On first run the model downloads (budget ~36GB of free disk: the kept download plus ollama's copy) and a benchmark runs to verify performance.

Expected performance​

Indicative, not a promise. A loaded machine is slower. Expect roughly 15-30 tok/s depending on chip, with Max and Ultra parts at the top of that range.

Tips​

  • Close other apps — the model shares unified memory with everything else. Safari with 50 tabs open means less RAM for inference.
  • Check Activity Monitor — look at the GPU tab to verify the model is using GPU, not CPU
  • 32GB is the floor — a 16GB or 24GB Mac can't run the text worker. Run a browser worker instead.
  • Keep it plugged in — macOS may throttle GPU performance on battery

Low performance?​

If you're getting significantly fewer tok/s than expected:

  1. Check free memory — close apps to free RAM
  2. Make sure you're on Apple Silicon, not Rosetta
  3. Run node -p "process.arch" — should say arm64, not x64
  4. See Troubleshooting