Apple Neural Engine leaderboard

Peak on-engine performance and fp16-correctness limits, measured with ANEForge across Apple Silicon.

Org Dataset Add your chip GitHub

All measurements

fp16 correctness cliffs

Magnitudes where the engine silently returns a wrong answer (clock-independent, so valid on battery).

Run it on your Mac (2 minutes, you get credited):
PYTHONPATH=. python3 bench/roofline_suite.py --contributor <your-gh-handle>
Then open a PR — your chip joins the board. See the roofline drive.