Apple’s M6 Is Its First 2nm Chip. Here’s What That Actually Changes

Apple's new M6 chip is here, and you may be wondering what's different this time around. It's a 2 nanometer chip, but process nodes are some of the most over-hyped numbers in tech. "2 nanometer" doesn't describe the size of anything you could measure on the die. Instead, it's a label for a manufacturing generation, and it has been marketing shorthand for years. So when Apple says M6 is its first 2nm chip, the useful question is "What did Apple do with that new headroom?"

Based on what Apple published today: two more CPU cores, two more GPU cores, a second Neural Engine, and more memory bandwidth, all packed into a smaller die. M6 debuts in the new Mac mini, which starts at $899 and ships September 22.

The CPU complex got restructured

M6 has a 12-core CPU split three ways: 2 super cores, 4 performance cores, and 6 efficiency cores. That three-tier arrangement is worth pausing on. The super cores handle single-threaded work, where Apple is claiming the fastest CPU core in the world. The performance cores draw less power and pitch in on multithreaded jobs. The efficiency cores absorb background tasks like indexing.

Why bother with three tiers instead of the usual two? Because the workload changed. An agentic AI task is not one long calculation. It's a burst of file reading, a compile, a model call, a tool call, then a wait, over and over. Apple built a CPU that can route each of those to different silicon. The claimed result is 1.2x faster multithreaded performance than M5 and 2.4x faster than M1.

Two Neural Engines instead of one

The Dual 16-core Neural Engine is the most concrete change in M6. Apple says it delivers up to double the peak compute of the previous generation, and that system frameworks can drive both engines simultaneously. Apps get faster model execution without anyone rewriting code, which is the rare kind of hardware upgrade that shows up for users immediately instead of after a two-year developer migration.

The GPU also grows to 12 cores, each with a Neural Accelerator, for what Apple calls nearly a 30 percent increase in peak GPU compute for AI over M5 and more than 8x over M1. In plain terms, that's the part that chews through your prompt before a local model starts answering.

Where the ceiling is

M6 supports up to 32GB of unified memory and 170GB/s of bandwidth. The bandwidth is up 10 percent from M5 and 2.5x from M1. The memory ceiling is the constraint that matters more. Apple spent this launch talking about running LLMs privately on device, and 32GB decides which models you can actually hold. Plenty of good ones fit. The biggest open-weight models do not.

There's also a graphics story that got buried under the AI talk: updated Dynamic Caching, a revised shader core, hardware-accelerated ray tracing, and 50 percent higher geometry rates. Apple points to Cyberpunk 2077 running twice as fast with ray tracing on M6 versus M4. Mac gaming remains a title-availability problem more than a silicon problem, but the silicon keeps getting less of an excuse.

The strategic read

Apple put its newest manufacturing node into its cheapest desktop rather than its most expensive one. The Mac Studio announced the same morning uses M5 Max and M5 Ultra dies, a generation behind on process. That ordering says Apple sees the volume opportunity in inexpensive, always-on machines that run AI locally, not in halo workstations.

One caveat to carry into every number above: these are Apple's own figures, measured on preproduction hardware using benchmarks Apple selected. Independent results land when reviewers get machines. Until then, the interesting claim is that Apple thinks the desk, not the data center, is where a lot of inference is headed.