We have official confirmation that Intel Cooper Lake is being re-positioned launching in 1H 2020 although only on the 4-socket and 8-socket Cedar Island
www.servethehome.com
This is the real scoop. Now stop spreading the fud without actually mentioning the real story or linking the news people can't read without subscription.
P.S. It is not Intel killing off 10nm server chips.
😉
from the article.
"....
- Intel remains on track for delivery of 10nm Ice Lake CPUs later this year.
...."
Yeah, illustrates that while Semiaccurate's "raw" info may be mostly good , the analysis ( "fitting info into big picture' , scope , ramifications ) are quite often Charlie seeing what he wants to see as opposed to what is really there. He'll probably still be beating the "10nm can't possibly work at all with any possible adjustment "drum well into 2021-22. ... in part because he loves saying "I told you so ... I was right on date xx.yy.zz " .
In the Xeon W space, there is not necessarily a good reason why Intel had to chase AMD into the double die , super large package zone. If Intel went with two 14 cores , 10nm dies they'd still be able to field something with 28 core range that the Mac Pro has now (and gets real work done for people) and keep the 10nm yield rates up. How the Xeon W Ice Lake are priced probably matters just as much to Apple as the the max core count available. If the yield held up could do 28 cores on single die like they do now and use about the same socket as they do now.
Two socket CooperLake was going to be a huge overlap with two socket ice Lake for a more than reasonable number of core counts. Neither one was probably going to top AMD's max core count per socket. But as been pointed out macOS isn't particularly a super high demand zone for absolute maximum core count as the sole primary selection driver. Cooper Lake int the two socket models was a hedge on not figuring out how to adjust well enough on the ice Lake iteration. If they don't need it because the adjust worked reasonably well then don't need that platform.
Intel probably will sell fewer server chips, but utter and complete collapse? ... probably not. They are probably going to step back from the sell everything to everybody approach. They still cover a large and broad spectrum of solutions.
The other factor is that these all these > 2 socket systems are shrinking.
https://www.nextplatform.com/2019/04/24/why-single-socket-servers-could-rule-the-future/
There is increasingly a narrow group that uses them. Google / FB / Baidu / etc that have low latency network data input ML inference workloads that need bfloat16 ... Intel is more than competitive with all of AMD's current offerings. Decent chance next ones too once weave in software stack optimizations. So Intel left the platform for the group that asked for it in the first place. Probably not going to loose most of those customers for that workload.
Did Intel ever think that Cooper Lake was going to be super competitive outside of the folks focused on bfloat16 . "Analysis" that presumes that seems a bit whacked. It was never designed in the first place to be primary option outside of that space for single/dual sockets workload since not much different than the cascade lake solutions.
