We have a thread here that's basically why can't Intel have 64cores like AMD?! Why doesn't AMD use monolithic?! Why is mobile monolithic?

This pops up too much so I'm answering and quote tweeting so that I can reference myself in the future when it invariably pops up again. https://twitter.com/nicafofita/status/1313583289693802497

The chiplet/MCM approach is a real part of why AMD is well positioned in the enterprise market today.

64C Rome with 8 CPU dies + I/O die is ~1000mm2. You cannot do that as a monolithic die, it well exceeds reticle limits (and yields would be awful even if this were viable).
1/?

Designing a series of monolithic dies to stretch from from 4-64C would be far more costly, and result in lower yields, and AMD still doesn't have anything like Intel's cashflow.
2/?

Would a monolithic die be cheaper/better for desktop? Yes, bit being able to re-use a single die from $100 4C consumer parts to $7000+ Enterprise chips is extremely cost effective, it also aids binning in being able to get the best chiplets for the purpose.
3/?

Intel has far more die variants then AMD and their only trying to scale to 28C on a die.
Their using two 28C dies at close to 700mm2 each to come close to 64C Epyc SKU's, and probably don't get great yields on the XCC Cascade Lake dies even using the very mature 14nm process
4/6

In the nearish future things will move past chiplet and onto interposers/stacked die solutions, which come with their own share of unique tradeoffs.
Genoa is almost certainly going to be using a stacked die I think, although this is moving off topic...
5/6

Renoir is monolithic because power draw is paramount on mobile. Also why less L3 then Matisse, SRAM chews power. Staying on chip is more efficient then driving external interfaces, IFOP links to the IOD matter. When they do advanced packaging things will change on mobile.
6/6