Apple’s latest silicon update is less about giving the Mac another routine performance bump and more about reshaping its chips around increasingly demanding AI workloads. The new M6 introduces Apple’s first 2-nanometer processor, while M5 Ultra takes the M-series into quad-die territory for the first time, with substantially more memory bandwidth and GPU compute.
The two chips target very different performance tiers. M6 is the next step for Apple’s mainstream silicon and debuts with a redesigned 12-core CPU comprising two “super” cores, four performance cores and six efficiency cores. That is two additional CPU cores compared with M5. Apple claims up to 1.2 times faster multithreaded performance than M5 and 2.4 times the performance of M1, although those numbers are based on the company’s own testing.
The move to a 2nm manufacturing process is arguably the bigger story. Smaller process nodes can allow greater transistor density while improving the balance between performance and power consumption, though real-world efficiency will depend on the finished implementation rather than process technology alone.
Apple is also devoting considerably more silicon to AI. M6 features a Dual 16-core Neural Engine, which the company says provides up to twice the peak compute of previous generations. System frameworks can use both engines together, while the chip’s 12-core GPU places a Neural Accelerator inside every GPU core.
Apple claims that combination provides nearly 30 percent more peak GPU AI compute than M5 and more than eight times the M1. The GPU also includes hardware-accelerated ray tracing, Dynamic Caching, an updated shader architecture and 50 percent higher geometry rates.
Memory bandwidth rises to 170GB/s, up 10 percent from M5, with support for up to 32GB of unified memory. M6 will first appear in the new Mac mini, but the architecture itself offers a clearer indication of where Apple is heading: more AI processing spread across dedicated neural hardware and the GPU rather than relying on one compute engine.
M5 Ultra takes that strategy much further. Apple is using its next-generation UltraFusion technology to connect two dual-die M5 Max chips, resulting in a four-die system that operates as one processor. Inter-die bandwidth exceeds 4.4TB/s, while the chip can be configured with up to a 36-core CPU and 80-core GPU.
That GPU again places Neural Accelerators in every core. Apple claims up to 4.5 times the peak GPU AI compute of M3 Ultra and more than six times that of M1 Ultra. Graphics performance is claimed to improve by as much as 40 percent over M3 Ultra, alongside third-generation hardware ray tracing and second-generation Dynamic Caching.
Perhaps M5 Ultra’s most consequential specification is memory. It supports as much as 512GB of unified memory with 1.2TB/s of bandwidth, 50 percent higher than M3 Ultra. That capacity opens the door to running exceptionally large AI models and datasets locally rather than depending entirely on remote accelerators. A 32-core Neural Engine and expanded media hardware round out the architecture.
M5 Ultra will power the new Mac Studio, but the machine is almost secondary to the silicon story. Apple is increasingly treating AI compute, memory capacity and bandwidth as fundamental processor specifications. M6 brings that philosophy to its next mainstream generation, while M5 Ultra shows just how far Apple is prepared to scale it.
