Gadgets · by Leila Karimi · 2026-10-12
M5 Macbooks are great but is anyone actually using the extra neural engine throughput for real work?
Upgraded mostly for battery life but ended up testing local LLM inference speeds too. The neural engine gains are real on paper but most local AI tools still don't fully leverage it compared to just brute forcing through unified memory + GPU cores. Curious if anyone's found tools that actually make good use of ANE for local inference yet.
0 sparks · 8 comments
Comments
Noah Miller · 0 likes
mlx is getting better at ANE utilization but still mostly GPU-bound for the big models, ANE shines more for smaller vision/audio models honestly
Maksym Bondar · 0 likes
battery life alone was worth the upgrade for me, getting a genuine full workday now on heavy compile jobs
Nima Ahmadi · 0 likes
same, i was skeptical of apple's claims but this time they actually delivered
Rapid Comet · 0 likes
whisper.cpp runs incredibly on ANE, if ur doing transcription work the m5 is a massive upgrade specifically for that
Bianca Santos · 0 likes
feels like apple silicon roadmap keeps outpacing software that actually uses the new silicon features, same story every generation
Noor Janssens · 0 likes
coreml conversion is still such a pain compared to just running pytorch/mlx directly, thats probably why ANE adoption lags
Beatriz Sousa · 0 likes
not upgrading until theres an actual reason beyond marginal gains, my m2 pro still handles everything i throw at it fine
Tomás Gómez · 0 likes
fair, i mostly upgraded bc my battery health was degrading not for raw perf