Apple M2 Die Shot and Architecture Analysis – Big Cost Increase And A15 Based IP
semianalysis.substack.com/p/apple-m2-die-shot-and-architecture
What Jamie wrote
The performance of the latest-and-greatest Apple Silicon, the new M2 chip, looks impressive to me. This article suggests however that Apple is showing strains and getting performance at a greater cost than should be expected.
The core itself is 21% larger than in M1, and 7% larger than A15. The big area of gen-on-gen growth is with the shared L2 cache which has gone from 12MB to 16MB compared to both M1 and A15. The AMX unit looks identical across A15 and M1 as well. The shared logic plane is also significantly larger which is is an indication that there is more bandwidth between the cores and the L2 cache and SLC. Overall, Apple spent 5.2mm2 on the big P-Cores, but the performance increase from them comes mostly from clock speeds.
This isn't the first time I've seen suggestions that the performance ramp isn't working on Apple Silicon. It is all conjecture though, so only time will tell. It is intriguing to me how they dissect the chip and think about the layout.