> My preliminary speed tests were fairly slow on my Macbook. However, once I deployed the app to an actual iPhone the speed of OCR was extremely promising (possibly due to the Vision framework using the GPU). I was then able to perform extremely accurate OCR on thousands of images in no time at all, even on the budget iPhone models like the 2nd gen SE.

He does mention running it on a macbook

I would guess that tests in this sentence refers to tests of the iOS app on the simulator.

Which would be slow expectedly

I would think it would run well on a M1/M2 Mac as a native app though, right?

That was my question as well. I'm wondering how much of a performance benefit the neural engine has on M series chips when compared to intel chips.

I would assume he's using an intel macbook and wouldn't have the gpu acceleration (and subsequent Vision framework integration) of the m1

There's ocrit, a CLI utility using Apple's Vision framework for OCR: https://github.com/insidegui/ocrit

What's the cost of building and running a cluster of iPhones vs Mac Minis?

in the article $40 second hand, imei banned and broken screen iPhones are being used so...

There's a ton of compute power available in the form of unused phones.