Semiaccurate (Demerjain ) generally has a quirk when it comes to Intel (and Nviidia). There really isn't a problem of Intel turning what will be called 10nm into something closer to what Global Foundries did with ( 14nm -> 12nm ). 10nm can work. Intel just has to undo some of the 5-6 things they piled on top of the original approach to 10nm. Instead of "tick tock" era where generally only try to change 1-2 major things at a time Intel did a bunch of them. Then preoceeded to pretend that stack 5-6 all on at the same time wasn't a principle part of the problem ( a bunch of upper and middle management "cover my butt' focus as oppose to fixing the problems. ). The whole "doomed" and it will never work stuff there is overblown at this point. Intel will get squeeze more at the profit margins but they can make this stuff for several products.
In the second anandtech link above there are
three phases to 10nm ( plain , + , and ++ ). Each of the '+" steps is a move to a slightly less dense implementation ( 'too high' of density being one of the root core problems of the initial 10nm process ). So those are heavily composed of just new design "libraries" for the same basic fab node ( there are likely also some incremental 'recipe' tweaks in there also , but not new infrastructure equipment and major changes in steps. Would need new 'masks' because mainly drawing something different at about the same resolution. )
I wouldn't bet on S-series missing all three of those. However, it very likely won't be Ice Lake. Willow Cove / Tiger Lake is a better candidate ( witch cache improvements which would likely make a significant contribution in S-class products).
https://www.anandtech.com/show/13699/intel-architecture-day-2018-core-future-hybrid-x86
Transistor optimizations would also have substantive impact at 10nm. The micro-architecture is making better use of what it has got ( not waiting on getting something better at the fab level ). Tiger Lake is probably can deploy on 10nm+ or ++ ( or perhaps 7nm if it arrives early ... which it probably won't). since a new micro-arch and next iteration it should get a "11th gen" tag ( if marketing isn't off high smoking something. )
Intel is going to make bigger dies with 10nm in 2020. Ice Lake SP is coming (Intel relatively doesn't have an option with the new EPYC pounding away at them ) . Doing a Willow lake S on 10nm+ (perhaps after some other products start on 10nm+ ) is probably doable and remove any "necessity" of waiting until 7nm to do something.
https://www.anandtech.com/show/1471...alable-in-new-socket-compatible-with-ice-lake
As for roadmap in semiaccurate article 3 .... as I commented earlier in the thread that is at least a 2018 (possibly 2017) dated slide. The desktop path may not have been mapped out at that point. Here is the tweakers desktop path for 2018-2020 (same timeline ) that showed up around the same time as that other mobile slide.
https://tweakers.net/nieuws/152112/...tot-eind-2020-bevat-geen-10nm-processors.html
No Ice lake Xeon SP processor at all. And yet as linked in above Intel's current roadmap slide have 10nm SP in 2020. Better contemporary data on what the yields are in late 2018 - early 2019 and finished up designs can do that. ( 10nm samples of Xeon SP Ice Lake are already being shipped out to large cloud players .... so it isn't like they are waiting to get to 'tape out' stage. ) Xeon SP has a many months long client validation phase that goes with it ( much longer then the mainstream consumer stuff. ) If the desktop roadmap is 'dates' the mobile one has a pretty good chance of being 'dated' also.
The mobile version has Ice Lake U (IL-U) coming out a quarter in front of Comet Lake U (CL-U) . So Intel introduces 10th Gen U and then 3-4 months later introduces 9th gen U ( with slower than the 10th gen GPU and lacking 10th gen feature set just spent 2 months heavily promoting ) . How looney tunes is that?
If there was a notion in that roadmap that 10th gen Ice Lake U would probably stumble and fall
behind the Comet Lake U roll out then that might make sense. 9th gen would ship in volume
before 10th gen dribbled in is highly limited quantities. to a small handful on low volume systems.
If they are making Comet lake U BGA (soldered "socket" ) compatible with 10th gen Ice Lake (and adding in I/O feature parity) that could make some sense in that it could be a drop in replacement for "missing" 10th gen IL-U . If the IL-U is just suppose to be the "Iris Pro" model of the 10th gen and the CL-U the affordable graphics of almost the same exact feature set then they both could be labeled as 10th as filling different roles.
But if CL-U is just primarily Whiskey Lake with a 50MHz clock bump .... why would it be taking so long. And why would they run Whiskey lake so long in overlap? The main part of CL-S is to crank up the core count. That isn't happening on the mobile roadmap at all. There is a "Q2 20 LP4x/DDR4 " note on it. ( which might be a variant with the originally targeted LPDDR4 updates that got stuck in Canon Lake / 10nm targets ).. That whole CL-U lane looks like a huge kludge. ( do everything in the U space.. 4-5 different versions and just pick which one is working later in 2018 approach )
Intel has put the graphics into the product name with 10th gen.
https://www.intel.com/content/www/us/en/processors/processor-numbers.html
If the CL-U has a different Gx than the Ice Lake ones ( G1, G4, G7) they have another set of SKU numbers to stuff them into.
If there is zero Ice Lake S coming then putting Comet lake S into 10th generation is a bit cheesy but it has some rationality to it. There is nothing there at 10nm in the 10th line up so Intel is filling it with 'something' . It do that (toss implementation generations
out the window as part of "generation" definition. ) at S level can also do that U level if find a rational to claim these are non overlapping parts and there is a "hole' to fill
if 'generation' is a group of products that have a common implementation feature set then Whiskey being decoupled from Coffee Lake Refresh was the dubious first stumble. Gen 9 should have been the coupling of the hardware Meltdown/Spectre/MDS fixes along with some other small tweaks.
Intel's "Emperor New Clothes" tactics here is part of the problem they need to fix. Instead of fun house mirrors gyrations on product naming they'd stop trying to cover things up with hocus pocus.
