Intel Corp (INTC)

[Unkwn]

Humm, makes sense, hope you are wrong though .. I think those chiplets with defects/binning problems will find their way around (avoid) the dumpster. They may have to accumulate for a good while. That 14nm is from GF makes for a willing source for tailored control hub, though packaging may add cost.

AMD could still use dies with defective cores/cache and combine them to a lower overall core count part, e.g. a 48 core variant or something, with the same amount of 7nm dies on the carrier. Whether this makes sense depends on many things, like yields (you need enough defective parts that are still usable) and platform costs for the Rome socket. Since it seems to be downward compatible to Naples, that shouldn't be an issue. The 48 core variant would be in between the current high end Naples with 32 cores and would also benefit from the 7nm improvements (and also the Zen 2 cores). They can further improve effective yield by separating the high and lower clocking parts for different SKUs, as well as for different power consuming parts (process corners). The Rome design gives a lot of options to AMD to really make use of the scrap.

I recently read an article about wafer costs between 28nm and 14nm Finfets. The wafer costs are said to be around 2000$ for 28nm compared to 7000$ for 14nm Finfets. This means that the costs are not compensated by the shrink, I think this gives you the trend. I would expect 7nm to be a lot more expensive than 14nm, maybe also not compensating for the gains by the shrink either. This means that the only reason to shrink further is performance gains, not economics, leaving 7nm designs for the high end for a while. Will be interesting how Intel's economics compare at 10nm, but I believe they'll have the same issue, therefore concentrating on the high margin/high end designs first, once their process yield is sufficient (for now, they have to experiment with smaller and lower clocked designs like Core M until they get the yields for performance parts).

A lot of speculation involved, but I wouldn't expect AMD to introduce Zen 2 based mainstream cores on 7nm first. Instead, they would stay at 12nm for mainstream Zen2 and introduce 7nm designs (Ryzen 7 and below) only when Intel forces them to do. We'll see if I am right with that assumption. A 7nm Ryzen 9 could be an option to fight back Intel's Core i9.

It's also interesting that AMD once mentioned that the 10nm (foundry) node is not of interest for them. It provides for better density, I know that from a comparison, so there must be other reasons for it. Could be performance or economics. 7nm seems to be a different animal though and that's most certainly not because of economics.