Recently, in a series of AI benchmark tests targeting large language models (LLMs), NVIDIA's mobile RTX 5090 graphics card demonstrated strong performance, significantly outpacing Apple's latest M5 Max chip in prompt processing and token generation speed, with a lead of up to 133%. However, the tests also revealed that when handling larger AI models or longer contexts, the advantages of Apple's unified memory architecture began to emerge.