iPhone chip's 50% memory bandwidth jump over A19 Pro matters more for local AI than 2nm
HankYeomans · x · 2026-09-10
Quoting PrinceCanuma, HankYeomans argues that everyone will cite the new iPhone's 2nm process, but the number that matters for local AI is memory bandwidth: roughly 50% more than the A19 Pro.
LLM decoding is bandwidth-bound — tokens/sec comes from moving weights faster, not from faster cores. He calls it the biggest single-generation bandwidth jump he's seen in an iPhone, with far more impact on on-device inference than the process node.
Related event: A20 Pro Memory Bandwidth Up 50%, Key for On-Device AI(2 posts)→
More from Infra
- Day 249 of GPU Programming: Tracking Cerebras From CS-1 to WSE-3 Turbo-Powered CS-4 — blaizedsouza · 2026-09-10
- GPT-6 Astra pre-training cost estimated at $432M, full model $1-2B — scaling01 · 2026-09-10
- Gensyn builds IR3DE-AXL, a decentralized collective inference network with no central gateway — benfielding · 2026-09-10
- Sam Altman: average person may burn 500B tokens a month within six years — rohanpaul_ai · 2026-09-10
- AI server demand is splitting into three ownership-based markets; ~1-1.5M on-prem servers need refresh — BenBajarin · 2026-09-10
- Marvell Ramps Supply Chain for AI Scale, Analyst Flags Substrates as the Bottleneck Few Can Master — BenBajarin · 2026-09-10