Prince Canuma(@Prince_Canuma):Everyone will quote 2nm. The number that matters for local AI is this one: 50% more memory bandwidth than A19 Pro. Decode is bandwidth-bound. You don't get 50% more tokens/sec from a faster core. You get it from moving weights faster. This is the biggest single-generation bandwidth jump I've seen in an iPhone.

Prince Canuma

@Prince_Canuma

Creator of (@Nativ_AI, mlx-audio & mlx-vlm) • working on something new Ex-@arcee_ai • @neptune_ai •

Joined July 2012

1.3K Following 22.9K Followers

Prince Canuma@Prince_Canuma

2026.09.09 18:14

Everyone will quote 2nm. The number that matters for local AI is this one: 50% more memory bandwidth than A19 Pro. Decode is bandwidth-bound. You don't get 50% more tokens/sec from a faster core. You get it from moving weights faster. This is the biggest single-generation bandwidth jump I've seen in an iPhone.