An iPhone 17 Pro Max Connected To An M4 Pro MacBook Pro Combined With Custom Software Resulted In Up To 44% Prefill Speeds When Running A 27B AI Model
Qwen3.8-27B is a capable AI model, assuming you have a sufficient combination of VRAM and system RAM. On the M4 Pro MacBook Pro, its 24GB of unified memory can be a limiting factor in prefill speeds when running the AI model. Fortunately, one user hooked up an iPhone 17 Pro Max to the portable Mac through a USB-C port, and by running custom software, he was able to get up to 44 percent prefill performance running the same AI model. Even with the memory raised to 20,480MB, the M4 Pro MacBook Pro is only limited to a 64K Context Window, […]
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.