Claude 5.5 视频 · @jurlycat
Bro borrowed his mom’s iPhone and turned it into a second GPU for his MacBook. Backburner runs Qwen 3.8 27B across a 24GB M4 Pro MacBook and an iPhone 17 Pro M
Bro borrowed his mom’s iPhone and turned it into a second GPU for his MacBook. Backburner runs Qwen 3.8 27B across a 24GB M4 Pro MacBook and an iPhone 17 Pro Max, connected by a 10Gb/s USB-C cable. How does a 27B model fit? It uses IQ4_XS quantization, roughly 4 bits per parameter. The Mac processes the first 40 layers of each batch. The iPhone handles the remaining 24 while the Mac starts the next batch. The developer reports 29–44% faster prefill, reducing the wait when an agent reads files or tool results. The phone also holds part of the KV cache for longer context. Each device keeps its own memory; the software splits the work. Built with Opus 5.5. Open source.