...
Regarding GPUs and custom designs: I think that the market has very clearly stated that it wants freedom to chose which GPU to use. Some needs nVidia for CUDA, end of story.
First I don't think that is the
whole market saying that. There is a highly vocal subset, they don't totally comprise the market. However, if Apple leaves an empty x16 slot in the upcoming Mac Pro then the freedom could be fullfilled. Fill the empty slot with an Nvidia card. Plug your 3rd party monitors into that card and you are done. If only wanted the Nvidia card for purely for compute ( ML inferencing , data calculations ) then can still use Apple's GPU just fine.
The major problem with the market "clearly stating" is that it completely ignores the Thunderbolt integration design constraints. If the solely PCI-e socket standard cards solved the problem that would be fine. They do
not. It is largely a form over function argument since that don't serve the function.
There are two parts to the tension that somewhat flow out of the 2013 Mac Pro design. One is Apple's push to control all of video output of every GPU card that is in the system. 100% of the video out flow has to go through TB or the world will end. Goofy ( in 2013 seem based on 'tail wags dog' notion that Mac Pro exist to drive Thunderbolt display docking station sales. Even more goofy now that Apple doesn't even have a display docking station product of their own. ). About as equally delusional is some user insistence that Apple has zero control over where at least some of the video output goes ( and boot screen compatibility ... ha only girly men look at boot screens ) . Both position are more about "control" than solving problem.
If a user wants to put a CUDA compute card in then simply providing a standards compliant slot settles that issue. Done. If Google started selling Tensor inference cards that fit in a standard slot and had macOS drivers. Again plop that compute card in and off to the races. Data capture card? Again into the slot and done. Could you do all three if there is just one x16 slot? No. However, just one slot will highly likely cover more than decent amount of the market. Can cover a bit more of that market with compute cards in TBv3 external enclosures (not everyone missing but some more. ) which the current Mac Pro can't really cover well (since restricted to TBv2).
And I don’t see Apple being committed to make a lot of custom GPUs for different needs, neither do I think the market will see any advantages to paying an Apple premium for those custom designs when they are used to just buy a PCI card and plug it in.
To the second part of the sentence, if the card is bundled with the system then really don't have a choice. Want to buy an Mac without buying macOS ... can't because it is bundled. The GPU bundled. Apple isn't going to sell GPU-less Macs any more than they are going to sell CPU-less Mac. You get to buy a complete system. Apple isn't a barebones shop or computer parts store. ( Why Nvidia can't seem to win any design bakeoffs of late has a good chance of being what Nvidia is doing at least as much as what Apple is doing. ). Apple's standard configuration GPU solution is going to be an engineered to Apple specification component. That isn't new; that is the way it has been even before the 2013 Mac Pro.
It is only really custom because the standard cards don't solve the problem (integration with TB). There is a somewhat defacto standard/guidelines for upper end laptop GPU cards (MXM ). Works great with TB integration (at least one iteration of the iMacs had them and thunderbolt before Apple fully integrated the GPUs. ). If the GPU and system vendors sat down and did a TB graphics card standard for "large" cards and Apple didn't follow it then yelping Apple for being custom would have some merit. There isn't so custom is what you get when there is no standard. (e.g., Apple moving the notch slightly on M.2 for custom SSDs is worth complaining about. Apple doing a custom SSD card before there was an finalized and approved M.2 standard wasn't/isn't.)
To the first part of the sentence, if Apple provides an empty slot then they don't have to make everything for everybody. As for diversity in the custom cards Apple could make for the upcoming Mac Pro that doesn't have to be all that narrow. The 27" iMac and the iMac Pro have GPU designs that could be relatively easily lifted and modified for the Mac Pro. So could have a sequence of minimal , good , better , and best ( typical apple standard config range)
tweaked top iMac GPU ("minimal") , tweaked iMac Pro GPU still down clocked ( "good" ) , up clocked iMac Pro (better) , and something unique ("best")
Only the last two would be a 100% Mac Pro R&D spend. The first two probably wouldn't cost as much as either one of those independent ones. If those are all AMD GPUs then the "AMD haters" can just buy the first option and fill the empty slot with something else ( more than likely cheaper than buying a TB external PCI-e enclosure box.)
There is a delusion that if simply nuked Apple from doing GPU cards that there would be this missive inrush of 3rd party Mac GPU card vendors. There likely won't be. There was no massive inrush of back in 2006-2009 time period. That was
before the MBP, iMac, and now iMac Pro skimmed off a decent number of Mac Pro user base. This new Mac Pro is going to have an even
smaller user base. Mostly likely what is going to happen is that there are going to be some "happen to work" PC cards that work better in the post boot status.
That would only make sense if Apple came up with a new, superior connection for GPUs ala nVidia’s NVLink, but now I’m dreaming. That is what an innovative Apple would do, IMO.
NVLink has nothing to do with the boot or TB interface issue. "Superior" isn't the hurdle here. One that actually works is the first step. Don't pump the DisplayPort output out of the box when it is needed in the box. That's it; simply function over form.
P.S. note that Apple is now a full fledged GPU vendor. the iOS devices are moving to fully custom Apple GPUs. With time this too could come to the Macs also. It is more a matter of how much additional resources Apple has to assign to more silicon design work and how well there GPU scales up. It would just be Apple is selecting AMD over Nvidia. It could get to put Apple is selecting Apple over all of them. Telling Apple they can't use Apple because it doesn't look like the form factor from 10 years ago is going to be tough. As long as can put a 2nd GPU into the system though it isn't a dead end or "painting into a corner" issue.