Skip to main content

macOS setup (Apple Silicon)

Requirements

  • Apple Silicon (M1, M2, M3, M4) — Intel Macs are not recommended (no Metal acceleration, very slow)
  • 16GB RAM minimum — the model needs ~10GB
  • Node.js 18+

Install Node.js

Using Homebrew:

brew install node

Or download from nodejs.org.

Run the worker

npx @c0mpute/worker --token <your-token>

That's it. node-llama-cpp automatically detects Metal on Apple Silicon — no extra drivers or configuration needed.

On first run, the model downloads (~9GB) and a benchmark runs to verify performance.

Expected performance

ChipExpected tok/s
M120-30
M1 Pro/Max30-45
M225-35
M2 Pro/Max35-50
M330-40
M3 Pro/Max50-70
M435-50
M4 Pro/Max60-80+

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
  • 16GB M1 works but you'll be tight on memory. 24GB+ is more comfortable.
  • 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