Thanks for the detailed info. That matches what I discovered on my own recently, and adds even more info. Essentially the conclusion I came to was that all TB is, is a DP protocol packet, and PCIe GPIO protocol packet multiplexor. With some extra logic to maintain the real time data rate needed by the DP protocol.
It is not. Thunderbolt is it own data protocol ( not DP , not PCI-e ) and to ease integration it appears to the host system as a PCI-e switch. The protocol can handle routing requirements for extended, external, daisy chains which PCie doesn't. Since it needs the very early boot aspects of a PCI-e switch (so can find devices ) and/or a displayPort output (so can see boot ) you have some GPIO ( along with some of the hot plug and play for those ).
The DP input from the motherboard has 100% relatively easy coverage from all x86 laptops ( both Intel and AMD) since they all have iGPUs. Same is true of > 50+% of all desktops ( the percentage floor might even be as high as 65% at this point. ). All have iGPUs and
In the end with standard PCIe GPUs there is no way to get TB output from these cards without custom connectors or extra cables.
Currently, statistically it is discrete PCIe GPUs that are odd. No, Thunderbolt is not optimally design around them. That is because they are disappearing from the personal computer market in terms of overall percentage. Is the percentage zero? No. Is it going to be zero anytime soon? No. But they are no where near the dominant driver of where the overall market is going.
And that means no Mac Pro will ever accept a standard PC GPU ever again, because of TB.
That is an over statement. As pointed out above the overall market is moving is a iGPU direction. Mac Pro is heading in that same direction. It is a huge miss the forest for the trees if try to posit that TB is the principal driver here. It isn't. Overall market trends is followed by Apple wanting to apply uniformity across the product line. Across Apple ... Everything is moving to Flash ( few corner cases where $/GB of Flash hasn't gotten "affordable enough") . Everything is dropping Optical Drives. etc.
Apple's primary need for GPU subsystem designs is for ones that have skills in delivering embedded solutions. It is simpler to develop an larger staff of experienced embedded GPU designers if all you do is embedded GPU design.
Well maybe as a second GPU add in, but a Mac Pro with a single GPU and TB means custom cards exclusively built by Apple.
The 2nd GPU is the "Compute" GPU. It isn't hooked to TB but also is not going to be the primary system design driver. If anything Apple under resourced putting resources behind the Compute GPU ( wider, deeper OpenCL support, ECC VRAM option , etc. )
Unless they come up with a new PCIe connector standard that includes DP out to the mother board.
There already was a external PCI-e standard before Thunderbolt was invented. Almost nobody uses it then or now. Thunderbolt was not invented to solve the general PCI-e external problem. Most of the
What I still haven't figured out is why is not possible to have a TB port that only does PCIe GPIO. That seems like a wholly artificial restriction, imposed by intel.
1. PCIe only TB doesn't nothing for widening the adoption of DisplayPort. ( the yet another connector is one aspect that represses external PCIe). In ramping up another 'new' port having other uses is an asset. ( If USB 3.0 was 100% incompatible with USB 2.0 it would not go as fast as it did. ) Why would DisplayPort agrees to a "You loose and I win" suggestion from Thunderbolt ?
[ whatever possible versus getting along with other people often leads to constraints. Everybody doesn't want the same thing. The larger the group the more true that is. ]
2. User confusion. Wait until some of the Type-C alternative modes are present on some systems and not on others. User will be confused because have cables physically compatible with the socket and it doesn't work. When someone plugs in a DisplayPort cable that fits, then it just works. That is a plus not a minus.
When get to USB Type C ports and how to point out to users that the tiny logo next to the port means you can plug in USB but not DisplayPort cable then will get "but the cable fits why doesn't it work?" complaints. USB will probably survive because has huge inertial behind it. Thunderbolt was starting from scratch. [ In the Type-C space, the TB probably probably will be the one port that covers all of the expected options ]
3. There already is an external PCIe standard. As the other system vendors who they don't put it on systems. Apple/Intel is probably will point to the same stuff. Never the less the box-with-slot zealots made a big stink about this. So Intel allowing some exceptions for systems with no iGPU, but it is by no means "saving" Thunderbolt in terms of response. [ And like HP card presents, there is a user expectation now that you will get DP output downstream and folks have set up configurations which depend upon video going dow n a TB network chain. ]
[doublepost=1470407172][/doublepost]
I agree with
@tralfaz that TB as it currently exists is a mess.
For laptops it is a mess? Not even close. The Mac Pro context is a narrow corner case.
Apple most certainly knows this, which may at least partially explain why there have been no updates on Mac hardware, with the exception the retina MacBook, for far too long.
Intel hiccups and users slowing their pace of new system buys has nothing to do with it.... cough... not.
I wonder if Apple plans to completely rid themselves of TB display ports and go with DP1.3 directly, maybe using USB-C?
Type-C is a just a port type. DP 1.3 and the DisplayPort phyical display part aren't connected either. TB v3 uses Type-C. Yes Apple will probably go with TB v3. Apple contributed 30-40% of the standards development team for Type-C (and a similar amount to TB ), so they probably are.
Apple might need a work on a more affordable adapter too. Initial TB3-TB2 adapters are probably going to have more than a few folks grumbling on price.
http://www.anandtech.com/show/10542...p-thunderbolt-3-to-thunderbolt-adapter-review
Apple may be trailing on doing USB 3.1 gen2. ( they dragged their feet on USB 3.0 deployment so it would not be surprising if dragged their feet on USB 3.1 gen2. ( gen1 is essentially just USB 3.)). Lagging driver development on Apple's part wouldn't be surprising. Throw on top eGPU support, and it would be even more likely lagging driver development effort.
