Yields on larger chips are still substantially worse than smaller chips, even with harvesting.
I'm genuinely having trouble understanding what you mean by this.
Rocket Lake being backported to 14nm means that 10nm can be allocated in greater proportion toward higher-priced chips like Alder Lake and Ice Lake SP. Seems like it would be good for production volume.
In laptop processors they can easily show efficiency gains from the 10nm process and IPC improvement, It appears most laptop processors end up running a power envelope lower then the ideal performance per watt efficiency. Server Processors with higher core counts means you can run more workloads per server, again providing efficiency gains. However desktop / gaming tends to be smaller core count + higher frequency with little concern of efficiency outside of quality of life factors (i.e. don't make me use a 1KW chiller). Intel has been pushing 5ghz processor frequency for years, and rocket lake continues that push (5.3ghz boost), when they drop frequency to move to 10nm, its hard to see an IPC improvement that is able to paper over that.
However alder lake CPUs will have a thread count advantage, so at least with 24 threads it should be able to show generational improvement over the current 8c/16 rocket lake parts. That will allow them to at least argue their value with select benchmarks and intel only features. Those 8 efficiency cores will likely be a BIG win on laptop, but on desktop I doubt they will compare favourably to the full fat cores on a current Ryzen 5900x (i.e. a currently available 24 core processor).
Intel is going to have at least 1 more BAD mainstream desktop generation before they can truely compete on the mainstream high end, however there is a chance they have something like a HEDT part that would allow them to at least save face. That being said, given a choice, Intel will give up desktop market share for the faster growing laptop and server markets.
I'm not seeing a good reason for thinking this is the case. Server CPUs are harder to fab (much larger die area) and they need to fab more of them (desktop CPUs are relatively niche compared to mobile and server CPUs).
If anything this is a sign that 10nm is fully ready.
Day by day, more data centers get AMD systems by choice or by requirement (Oh, you want 8xA100 nVidia made modules with maximum performance. You need an AMD CPU since it has more PCI lanes, for example).
You don't see much AMD server CPUs around because first generation and most of the second generation has completely bought by FAANG, Dropbox, et al.
As the productions ramps up with newer generation, we can buy the overflowing parts after most of the production is gobbled up by these buyers.
When you configure the system full-out with GPUs & HBAs, the number of lanes becomes a matter of necessity rather than a spec which you drool over.
A PCIe lane is a PCIe lane. Its capacity, latency and speed is fixed, and you need these with minimum number of PCIe switches to saturate the devices and servers you have, at least in our scenario.
All of the other benchmarks in the PCMark test suite push the bottleneck down to the storage device.
One would think Intel might want to build a storage array that could stress the PCIe lanes but then that might show an entirely different picture than the one Intel is portraying.
PCIe efficiency is such buzz-none-word.
once you get those easy wins the rest take a bit to chip away at.
while initial growth was 100x, that pace is going to slow down as marketshare increases.