iPhone 17 pro max enhances macbook pro AI performance by 44%

A Reddit user has demonstrated a method to significantly boost the pre-filling speed of the Qwen3.8-27B AI model when running on a MacBook Pro with an M4 Pro chip. By connecting the device to an iPhone 17 Pro Max via USB-C and using custom software, the user achieved a 44% performance increase. The setup splits computational tasks between the Mac and the iPhone, with the Mac handling the first 40 layers of token processing and the iPhone’s A19 Pro GPU managing the remaining layers. This collaboration reduced token write times and improved pre-filling speeds, especially at 16K context length. The project, named 'backburner,' is now open-source on GitHub. While the method shows promise, it still has limitations, such as not accelerating text generation beyond 64K context. The upcoming A20 Pro chip in the iPhone 18 series is expected to offer even greater performance potential.