Apple debuts M6 on 2 nm and M5 Ultra with quad-die architecture

苹果推出 M6 和 M5 Ultra，性能和 AI 算力大幅跃升

Apple put its first 2 nm desktop chip, M6, into the new Mac mini. It bumps both CPU and GPU to 12 cores, doubles the Neural Engine to dual 16-core blocks, and raises unified memory bandwidth to 170 GB/s. Single-thread performance is the fastest in the world; multi-thread is 1.2x over M5, and GPU AI peak compute is up nearly 30%. Alongside it, M5 Ultra uses a quad-die package for up to a 36-core CPU and 80-core GPU with 1.2 TB/s memory bandwidth—50% more than M3 Ultra—aimed at running large models locally and heavy pro workflows. Both chips emphasize performance per watt; M6 supports up to 32 GB unified memory. The post does not disclose pricing or exact availability dates.

Why it matters: Apple drops M6 and M5 Ultra with first 2nm process and quad-die packaging, plus concrete AI compute gains. The ding is that this is a press release — no third-party benchmarks yet, so real-world local inference performance is still TBD.