December 10th, still not available.
Quoting
The Register, according to HP, shipments are going to supercomputer / hyperscale systems first, then servers, workstations, desktops in that order. Presumably "most favored nation" vendors like HP first, then white-box integrators, with anything left falling into the retail channel.
Intel really blew it with their 10nm fiasco, and not realizing soon enough that there was a problem and looking for alternate solutions. The pipeline for wafers used to be pretty long, so even with a perfect fab process and high yields appearing by magic today, you wouldn't see finished parts coming out of Intel for quite a while.
If I were Intel, I'd probably OTP any dies that met the server specs as server, whereas when times were flush some server-spec parts were fused as workstation parts and sold that way would now be sold as server parts. And do the same all the way down the line. This assumes that various parts from a specific generation are OTP'd to the various market segments based on binning and market demand. I can't see Intel having a zillion different production lines (or worse, split masks) - I'd guess there are at most 6 different base parts being produced for any specific generation, then binned as needed.
Also, assuming that I required (these are numbers I randomly pulled out of an orifice, I don't know anything about current binning processes and have never worked for Intel) a 25% margin (part must be clockable at 25% above the speed the part is fused and sold as) and similar reductions in other margining tests, I might drop that margin to 20% and issue a "Confidential" erratum to my OEM customers saying that these parts still meet specs, but that "temporary process changes" were made for the following parts in order to meet production goals.