Hacker News Reader

I don't if Intel is doing the same, but AMD's chiplet-based Zen 2 and Zen 3 processors use a less dense manufacturing process (12nm) for their IO-die than for their CPU-die (7nm), which results in the IO-die being comparable large.

The uncore has been growing every generation so this is not new. I've wondered why it's so large myself. Intel has the audio DSP, ISP, GNA, and now VPU that mostly aren't used so maybe they account for the bloat.

I'm not a semiconductor engineer so I could be totally off here but.

My guess is that for AMD and likely Intel the reason the IO die is so large is because of all the PHY's on it (DDR, USB, etc) which would give a minimum size of the IO die for chiplets.

Which if true explains why you wouldn't care as much about the process node for the IO chiplet as it has a minimum physical size anyway.

iGPU, video de/encoder, USB4, I'm not surprised it's larger.

The GPU is on its own tile