| L1 instruction cache | L1 data cache | L2 cache | Minimum clock | Maximum clock | |
|---|---|---|---|---|---|
| M6 electronic core | 160KB | 128KB | 8 MB shared between a 6-core cluster | 1,010MHz | 2.908MHz |
| P M6 core | 128KB | 64KB | 20 MB split between P and S cores | 1,440MHz | 4.788MHz |
| M6 super core | 192KB | 96KB | 20 MB split between P and S cores | 1,440MHz | 4.788MHz |
The M6 gets additional GPU cores for the first time since the M2, increasing the core count from 10 to 12. These cores include improved versions of the integrated neural accelerators that Apple introduced in the M5 series GPU to accelerate AI and machine learning workloads, such as ray tracing and MetalFX upscaling. A “dual 16-core neural engine” also increases the number of neural engine cores from 16 to 32. Saying you have 2×16 of something seems like a roundabout way of saying you have 32 of something to me, but I guess the end result is the same either way.
A small detail on the M6 variants used by the Mac mini: the base 16GB Mac mini offers 153 GB/s of memory bandwidth, the same amount as the M5. The 24GB and 32GB versions offer 170GB/s, an 11 percent increase that should marginally improve the graphics performance and speed of any local AI model you run on this machine. Our review unit is the 24GB version.
What about the M6 Pro, M6 Max or M6 Ultra? If the rumors are true, the M6 could end up being a solitary chip.
Bloomberg’s Mark Gurman reports that Apple is not planning a full generation of M6 chips and could move directly to the M7 brand for the Pro, Max and Ultra chips. It would be a first for the company, which skipped shipping an M4 Ultra, but so far has shipped a basic chip, from the Pro and Max series in each generation.
But it makes sense for a couple of reasons. The first is that most Some of the M6’s gains can be attributed to its additional CPU and GPU cores relative to the M5, and the M5 Pro, M5 Max, and M5 Ultra obviously still have more of both types of cores than the M6. The second is that it’s still pretty early for TSMC’s 2nm process, after three generations of Apple using versions of TSMC’s 3nm process, and Apple is probably fighting every other fabless chipmaker on Earth to reserve capacity. Apple could stick with smaller, simpler 2nm chips for now and save its larger, more complex chips for later, when capacity is higher, yields are better and costs can be lower.
