Nothing I said is inconsistent with deconstruct60's theory on a 22xx/32xx split in the Xeon-W lineup - as a matter of fact, it matches very well.
22xx = iMac Pro (up to 18 cores, ~2000 pin socket)
32xx = Mac Pro (up to 28 (32?) cores, ~3500 pin socket).
You are trying to move the goal posts. Almost nothing you are proposing is consistence with what I have expressed.
The 32xx isn't an "up to 28". It is a " greater than 18 and up to 28" product. There is almost zero rational for Intel to overlap the 32xx with core counts that match the 22xx series. They would
four product lines that directly overlapped using the same underlying die. Some Core i9's , W 22xx , W 32xx , and SP . There is only so much hocus pocus Intel can slap on these exact same die. over half of the W 32xx line up overlapping with the W 22xx line up doesn't to jack squat to help with marketing against Threadripper. This all mostly some highly contrived stuff to motivate putting the Mac Pro solely into some "largest socket" positioning relative to the iMac Pro. that is just deeply misguided in driving Mac Pro costs up just be high for the sake of being higher. Same "Low Core Count" die in a larger package would primarily just cost more. ( Apple is highly unlikely going to slap an unusually high number of DIMMs into the system. Or want to throw volume at bigger sockets because they are big. )
Whether any of the 32xx series are capable of dual-socket configurations is up in the air,
That isn't up in the air at all. The whole point of the W series is that they are one socket solutions. That it. The same way the E5 1600 ( and 3500-3600 series before were different from the E5 2600 ( 5500-5600 ). The whole W series is one socket and will be one socket.
The 32xx will probably stick around because the upper limits on core count just takes more die space (and package space to hold all the die(s) ).
Initially the primary point was to couple the W series with a PCH that is closer to the relatively faster moving desktop PCH. The 31xx (and likely 32xx) are entirely off the road and in the swamp with respect to that. It will be hard coupled to the SP series cadence. The one upside that Intel may bring in with the W 32xx series is enabling the Optane DIMM capability ( and restricting that from the i9's and W 22xx ). [ As I said before it is highly doubtful Apple is going to want to touch the Optane DIMMs at all and will be far more interested in more affordable 22xx then chasing super sockets. ]
If Intel brands them this way, then the minimum core count of the Mac Pro will be set by the minimum count available on the big socket (again, I agree with deconstruct60 - Apple isn't going to do two motherboards in the same machine). For precisely this reason (not just Apple, but other workstation vendors as well), I expect Intel to provide some level of overlap - we won't see ~2000 pin sockets go up to 18 cores, and ~3500 pin sockets only over 18 cores.
Yes on not Apple not making two motherboards (and different PCHs for Mac Pro). No on Intel "has to" do sub 20 core 32xx series. Intel has about zero need to provide that and also probably about zero system vendors asking for it.
The i9s and W 22xx already overlap. System vendors don't need 3 overlappers. Most system vendors are quite happy in selling the SP in top end workstation (with empty sockets ... not that Apple would go that way. )
The 3175x and probably 32xx are primarily just bumps in the road. The most of the W series is using the Low Core count (LCC) and High core count (HCC) dies.
https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/5
If Intel gets a 20% shrink out of going from 14nm to 10nm ( those number alone are 29% shrink ) then Intel will be able to bump those core counts to 14 ( for LCC) and 22 (HCC). For example, the HCC has 5 rows of cores. One of those rows represents 20% of that core space. So if shrink all of them 20% then will have room for another row. ( the uncore parts could shrink to , but there is stuff to add there too like inter-die fabric to be used in a chiplet style building of multichip modules (MCM) ) .
They might not even need the HCC and XCC at that point. If the LCC came with 14 cores than two LCCs would be 28. And three LCC would be 42. the W 2xxx could/would take up to two dies. The W 3xxx could/would take up to 3 dies. The primarily point being that Intel is about one die shrink away from putting 22-28 cores in the 2xxx series. At some point that kind of core count is "enough" for wide range of folks.
There is a much smaller group that are on the high core , but lower clock chase. If Apple can just buy the medium size MCM modules with the die(s) already packaged up then that is largely sufficient. Incremental fab progress will probably more core count coverage over time.
It's possible that Intel offers the big socket all the way down - that there might be 32xx options with 10, 8 or even 6 cores?
Possible; just not very probable. Intel needs yet another overlap product they they need another hole in the head. They could drill another hole in the head but it more likely it won't buy them much beside a deeper hole they have dug themselves into.
What Intel needs to get their fab rollout together. Not yet another Emperor new clothes product.
There are Xeon-SP options without a ton of cores, so it wouldn't be that hard for Intel to strip a few features out and call them Xeon-W 32xx.
Those are some increasingly wonky products. The super low core count, but high clock stuff usually are targeted at the virtual networking and commnications servers. The new SP 2nd generation products have updates that are more customized for those
"...
- Y = Speed Select Models ...
- N = Networking/NFV Specialized
..."
Speed Select allow 2-4 cores to be tagged with a relatively higher clock base while 4-6 have a low clock base. If the latency nexus of the virtual network switching is assigned/pinned to those 'selected' cores and the rest of the 'overhead" system is relegated to the lows then don't need those binned very high clock very low count SP processors as much anymore. Not all of those targeted customers are going to buy into that so the low count ones are still around this iteration. But if those are successful ... the low count stuff will probably disappear out of the SP line up. If the Xeon D series gets a decent upgrade on the next shrink even more likely. ( also the Xeon E line up covers up to 6 now and may go to 10 by around early 2020. again how many 4-10 edge server Xeon CPUs does Intel really need when based on same microarchitecture???? ).
Yes the workstation vendors do have "low ball" models with these extra low count SP processors (with an empty CPU socket to limbo even lower), but it is extremely doubtful Apple is even remotely interested in chasing those kinds of buyers. Apple has about zero interest in folks who are looking to buy empty CPU sockets. Macs deployed into virtual networking devices .. again almost zero. Macs sold as effectively barebones boxes ... again almost zero.
If they do that, I'd expect that Apple might ignore those chips - just as they do with a few low core count variants of the Xeon-W that don't show up in the iMac Pro since they'd overlap with the iMac. My guess is that Apple starts the Mac Pro at 12 cores if Intel makes that chip available. If Intel starts at 14 cores, so will Apple.
Nope on the latter part. The notion that the Mac Pro can't overlap the iMac Pro is deeply unmotivated. It makes about zero sense. Apple didn't withdraw the iMac Pro 8 core model when the iMac got to 8 cores. The iMac 2013 model had 4 cores that overlapped with the Mac Pro 4 cores. This whole core count domination mania is largely contrived, because Apple really hasn't followed that as a dogma in the past at all.
A Mac Pro topping out at 18 cores would be 10 more ( more than 100%) than the top end BTO iMac 2018. Even if iMacs in 2020 bump to 10 cores that will still be somewhat close to almost 100% more.
The Mac Pro gapping the iMac Pro came be very substantively done with other features than raw core count. If Apple was absolutely fanatical about core count they could just switch to AMD. Apple painting themselves into the corner of only the upper end of Intel W and/or SP product lines is just as fundamentally flawed as the corner they painted themselves into with the Mac Pro 2013. Those subsets of Intel's product line have increasingly noncompetitive aspecs to them in the single user system space. They aren't designed for that. They make far more sense for workloads of several users over which the costs can be spread.
What will Apple do if Intel starts the 32xx series at 18 or more cores? I suspect they'll get custom chips - they have enough leverage with Intel from their high-volume laptop and iMac business, not to mention iPhone modems!
Or far more likely.... just not buy them at all. They could have bigger leverage in
combining Mac Pro and iMac Intel W 2xxx series processors.
The Intel modem leverage is about worth crap now that Apple is spending 100's of millions to jump into the modem space themselves. Utter garbage leverage. Intel knows it is a game of music chairs and Apple is both playing and controlling when the music stops. Intel is mostly only worried about who to find to flush out the loss when Apple dumps them. [ Apparently according to another macrumors story Apple has "wafer first start" priority over just about all Intel products for Apple modems. That right there is a problem. Fixed 14nm capacity and Apple hogging the road ... but dialing back from their table pounding projected demands for a product that was suppose to "make it up" in volume. Yeah right ... even
less leverage. ]
If Apple chooses to go down the 32xx rabbit hole ( a huge mistake IMHO) then they will get chips that do a subset of what the normal 32xx series do. Intel will perhaps adjust the knobs on what is there ( tweak L3 cache balance and/or clocks ), but that is about it.
They may disappear in the workstation business, but they're a huge Intel customer overall, and they buy a lot of chips that have decent margins.
If Intel doesn't get their act together consistently then they won't be a huge Intel customer overall in a couple of years.
If Apple prunes off the bottom end of the Mac laptop line up for their own processor then Apple should probably put more weight into the narrower set of Intel products they are buying. Diluting their buying power over a wider set of products ( where they would be the more relatively smaller customer for that specific subset ) isn't going to get them more clout than they have now. ( Apple will probably yank out some of the higher volume (relative to the higher desktop Mac buys ) stuff first. "old' MBA price point (replace with iBook/ARM Macbook one port wonder) , edu-entry iMac non retina (if keep it around as non Retina). )
If Intel gets on track what Apple will be buying is what they don't want to do ( often in part because it doesn't have enough volume for them to justify doing it for themselves. ). If Intel stumbles and AMD doesn't then Apple can take their volume to AMD for which even a reduced volume would be a bigger deal to AMD.
