iPhone 17 Pro Max used as an external GPU for a MacBook Pro M4

Reddit user StayLameBro turned an iPhone 17 Pro Max into an external GPU for a MacBook Pro M4, splitting the workload of the Qwen 3.8 27B AI model between the two devices. Prefilling a 2,000-token file is 29–44% faster than the Mac alone, and a 27,000-token session gains 36–46%. The setup needs an iPhone 15 Pro or newer and a 10Gbps USB-C cable, with code on GitHub.

• Mac handles the first 40 layers of each token batch, iPhone GPU the rest
• Prefilling 2,000 tokens is 29% to 44% faster than the Mac alone
• A 27,000-token session gains 36–46% depending on the llama.cpp build
• Requires an iPhone 15 Pro or newer and a 10Gbps USB-C cable