.... 4 TB ports should be quite enough for most people I guess. Of course there always those who need more, but that should account for a small percentage of real users.
It is really need for some combination of TB and mDP ports. There are 6 on the current Mac Pro as much for "Thunderbolt" needs as for "connect my legacy monitor" needs. Instead of having the HMDI share a video source with the a pair of TB ports, just letting the HDMI port solely have the source doesn't really cut down the number of video sources feeds. HDMI connectors on monitors is at least as generally common at this point as mDP/DP ones. There is only a very small corner case for 6 monitors (or three first gen 4K ones ) that need mDP.
The USB 3 ports are already there to be used on the chipset, most of it is under used, even the GbE is not used - Apple opted for the extra Broadcom chip.
The GbE on the chipset ( C600 or C610 ) series actually consumes a PCIe link to physically hook up. So whether use Broadcom/etc GbE controller or Intel's marginally cheaper chipset-to-phys adapter it is still costing x1 PCIe v2 link. I think to do dual ports, it is just more straightforward to do two independent x1 GbE controllers.
USB 3 ports are on the C610 chipset ( for v3 or v4 ) but so far Apple isn't in a hurry to get there. It is a slippery slope of letting Apple slide back into the lazy mode of Intel USB controller drivers are the only ones they need to work on.
Broadwell is a bit of a mess, ...
...
Out of curiosity, M370X on rMBP is Cape Verde all over again,
Broadwell in and of itself wasn't really a mess. The transition to 14nm was the core problem. Intel's 14nm Atom line ( that isn't the Broadwell design) didn't ship on time either. The specifics of the microarchitecture independent of process probably where not the root cause issue.
The M370X as Cape Verde is somewhat indicative of the same problem. Both AMD and Nvidia are largely stuck on 28nm. Some of the rebrand volume is because there is no really new microarchitecture to move to. Getting more "optimize the current generation design" output. Relatively small tweaks to better run on the process tech currently holding at then. [ Nvidia had done a bit of 'tick tock' move with Maxwell is making more of 28nm. AMD doesn't run as broad or deep of a design pipeline. ]
The jumps between process level 22nm -> 14 nm -> 10nm are taking longer in part because it is getting substantially harder. Therefore harder to run a wide range of design teams in deep pipeline fashion.
GCN 1.0 rebrand as usual, old tech. What gives, Apple?
CPU 'older' ... GPU 'older' similar to the Mac Pro 2012. Probably a placeholder design. "what is easiest to do while spending most work on the more substantive upgrade".
Although, if the 5K video support is one cable ( DPv1.3 ) then it isn't so much old as tweaked. The revision number ( not model number) is more significant. If tweaked the display support it wouldn't be hard to tweak some of the missing OpenCL 2.0 support issues also (non working GCN 1.1 elements switched on and working in this revision). Why reuse the device id at that point perhaps has to do with keep driver complexity down.
If a two cable 5K solution and better die binning (higher clocking with lower power ) then just relatively old GPU design.
Maybe it's just a short term refresh while SkyLake is unavailable, and maybe next year there's a new model with a newer GPU, but that doesn't make a lot of sense.
Intel seems on track to do Gen 6 ( Skylake) quads first so probably a new MBP 15" in the late Fall. (The non Retina iMacs are now languishing too. So they'd be part of the Fall dog-and-pony show too ) Maybe AMD has uncorked their fab problems by then. Maybe Apple has squeezed better prices out of Nvidia (and/or they moved forward) by then.
