Apple opened preorders August 25 for the M6 Mac mini and M5 Ultra Mac Studio. Most configurations are due September 22; the Studio’s 512GB memory option is due in late October.
The M6 mini tops out at 32GB of shared memory, the pool available to both processor and graphics cores. Between it and the Studio is a $1,699 M5 Pro mini with up to 64GB; the Studio begins at 96GB.
What changes in the chips
M6 raises memory bandwidth from the M4 mini’s 120GB per second to as much as 170GB, an increase of about 40%. Each of its 12 graphics cores includes a Neural Accelerator, hardware built for the calculations AI models use.
The M5 Ultra combines four pieces of silicon and extends neural acceleration across as many as 80 graphics cores. Its memory can carry 1.2TB per second, 50% more than M3 Ultra, while maximum capacity stays at 512GB.
The systems do not ship until September 22, so no independent test yet establishes Apple’s claimed AI gains or a launch post’s claim that the 64GB M5 Pro can usefully run a compressed model with 70 billion learned settings.
Apple has already shown four Macs splitting a single AI model that was too large for one machine. The new Ultra does not raise the memory ceiling on any one box, but its faster memory and AI accelerators could make that kind of cluster a more capable way for teams to run very large models outside remote data centers.