Bluesky Thread

iPhone Air has an A19 Pro chip, which is has native matmul

September 10, 2025 View original thread

iPhone Air has an A19 Pro chip, which is has native matmul

full NVIDIA inference speed in laptops, maybe even phones too?

news.ycombinator.com/item?id=4518...

It has A19 Pro. A19 Pro has matmul acceleration in its GPU, the equivalent of Nvidia's Tensor cores.
This would make future Macs extremely viable for local LLMs. Currently, Macs have high memory bandwidth and high VRAM capacity but low prompt processing speeds. Give it a large context and it'll take forever before the first token is generated.
If the M5 generation gets this GPU upgrade, which I don't see why not, then the era of viable local LLM inferencing is upon us.
That's the most exciting thing from this Apple's event in my opinion.
PS. I also like the idea of the ultra thin iPhone Air, the 2x better noise cancellation and live translation of Airpods 3, high blood pressure detection of the new Watch, and the bold sexy orange color of the iPhone 17 Pro. Overall, this is as good as it gets for incremental updates in Apple's ecosystem in a while.

23 4

2 hours later

makes you wonder if Apple's been doing AI right all along

they kinda botched LLMs on a number of fronts, but the one thing AI will definitely need regardless of trends, hype & architecture is fast inference & matmul

they really nailed that for consumer devices

More like this