// Hacker Noon · 26 May 2026

I Made My iPhone's Neural Engine and GPU Run Inference Together as an Experiment. It Got Slower.

I made the iPhone GPU and Neural Engine run inference together, expecting 2x speed. It got slower instead, around 10% slower for a light model and almost 4x slower for a heavy one. On Apple Silicon every accelerator shares one memory bus, so two of them running together just fight for bandwidth and...

Hacker Noon

@hacker-noon · NSIbragim

hackernoon.com

Hacker Noon@hacker-noon