Given Apple's track record over the last decade if they did a redesign around Xeon E5 v4 ( Broadwell-EP ) now in 2016 I highly doubt they would come back in late 2017 or even early 2018 with anything that required a significant R&D work. Yes, Skylake-W could be a much better match to the Mac Pro design objectives, but if Apple hides in the hole for another year the Mac Pro market probably would be too severely damaged to be viable over the long term.
...
Imagine having full width GPUs as now, 2 or 3 TB3 controllers and 1 or 2 SSDs out of the CPU.
An optional 3D XPoint / Optane SSD (as caching and/or VRAM backing store ) and three TB v3 would probably work pretty well. If Optane pans out as advertised could be simpler to just use one of these instead of the RAID 0 configuration of the Radeon Pro SSD. Since not coming until late 2017-2018 the Optane would have a higher chance at easy availability ( and more proven track record.). A single even more low latency, 'M.2 sized' SSD wouldn't need a PCI-e switch (besides the one embedded in CPU ) and more space effective than two SSDs. [ In contrast to the Radeon Pro SSD design] It would need some more macOS updates if using as specialized storage (Fusion like cache or as VRAM backing store mapped files).
There may also be some "low cost" 10GbE option that the "4th" x4 lane could be allocated to. Skylake-EP is getting some 10GbE options. If those get delivered via some relatively "low cost" PHYs implementation ( similar to current two chip solution PCH+PHYs for 1GbE ) that could be very easy path to brining Mac Pro Ethernet into the 21st century. [Desktop Systems that target extreme bulk storage on network should be greatly helped with low cost 10GbE availability. ]
Either Optane or Intel special PHYs means more bundled stuff for Intel to sell, so I suspect support for one or both will be made attractive to Apple to include.
No bottlenecks even from DMI3.
As the PCH loads up on the "up to 20 PCIe 3.0 " then bottleneck go up. It is more a physical lane expander in the 3.0 space than the step up Skylane-W takes by going from 40 to 48 lanes ( presuming there is respective step up in internal network inside the CPU. )
I expect the default SSDs will still go here. Unless, Apple gives a bit on miniizing volume don't suspect will get two general usage SSDs.
[doublepost=1469986080][/doublepost]
Or maybe -W will use the same chipset as the new enthusiast -X processors, Basin Falls-X, with even better specs, 24 PCIe lanes:
Quite likely since do that now. ( 610 works for both -E and Xeon E5 1600 v3-v4 ). Most of what is happening at Skylake is a mapping from -E and -EP to -K and -W for same family of product. [ But of course muddled and bound to fuel confusion by Intel marketing by slapping an exception for the i7 x700 series in there with the -K to -X mapping. ]
But overall yes. As the desktop market slows and matures, it is going to be more cost effective to have one PCH for all mid-to-high end desktops ( with internal features flip on/off for market differentiation) .
Looks like this is could be a "rob Peter to pay Paul" exercise.
Skylake-W 48 lanes. Skylake-X 44 lanes ( -4 )
PCH 20 lanes PCH 24 lanes ( +4)
Create a PCH chip with an optional connect of x4 to drive the SATA ( they have done it before for SAS ):
http://www.intel.com/content/www/us/en/chipsets/server-chipsets/c600-chipset-diagram.html
Other viable candidate is just more Flex IO. Can drive up the PCI-e v3 lanes a bit higher by driving down SATA lanes. Same pins coming out just allocation mix changed.
