Understand
Apple M6 vs M5 Chip Deep-Dive: Die Analysis and Architectural Differences
Overview
A detailed silicon floorplan and die analysis comparing Apple's M5 and M6 chips, based on findings from the SemiAnalysis STEEL team. The Apple M5, built on TSMC's N3P node, features a 154 sq mm die with an 8-core S-core setup, 10-core GPU, and 16MB System Level Cache. In contrast, the M6 is produced on TSMC's N2 process node using Gate-All-Around transistors, measuring 141.6 sq mm. The M6 introduces a redesigned combined S- and P-core cluster (replacing two S-cores with four new P-cores), an upgraded 12-core GPU, a doubled dual-16-core Neural Engine, and expanded cache capacities.
Key points
- The Apple M5 is manufactured on TSMC's N3P node with a die size of approximately 154 square millimeters, featuring an 8-core S-core configuration, 10-core GPU, and a 16MB System Level Cache.
- The Apple M6 is built on TSMC's new N2 process node with Gate-All-Around transistors, resulting in an 8% smaller die size of 141.6 square millimeters.
- The M6 upgrades to a 12-core GPU, marking the first core-count increase for base M-series GPUs since the M2.
- The M6 features a combined S- and P-core cluster, replacing two large S-cores with four smaller P-cores, alongside larger L2 cache slices (1.5MB per S-core and 20MB shared L2 cache).
- The Apple Neural Engine (ANE) on the M6 is significantly updated with a dual 16-core NPU design for enhanced on-device model and Apple Intelligence performance.
- Both chips maintain a 128-bit memory interface using LPDDR5X memory PHYs, while the M6 upgrades its PCI-Express configuration to six uniform PCIe gen4 lanes.