And here we go again to the PCIe slots, the mMP will include some sort of PCIe expansion beyond TB3 (PCIe X4 is far enough for most applications), even I doubt Apple to allow any STD GPUs it may offer a way to access PCIe lines (as on a propietaty licensed Made for Mac Pro alike program) or even an adapter to hold STD PCIe peripheral instead one GPU, but Apple will enforce to boot only apple-blesed GPUs (the golden eggs hen for WS Market ), Apple has N ways to enforce it not just controlling the Slot Form factor (STD PCIe is has no DP1.4 feedback lines to feed TB3 headers).
Apple isn't that crazy. The above is just hand waving. If Apple made room in volume , power , and cooling for an internal PCI-e card then they'd just use a standard slot. Making gyrations of some Rube Goldberg adapter doesn't do anything productive either for Apple
or the end users. It is a loose/loose design option.
The restriction Apple probably place is on a generic standard PCI-e GPU card being placed in the slot where the Apple boot/default GPU goes. It is extremely likely that restriction would involve being coupled to Thunderbolt and the standard cards do not have that functionality; so they shouldn't fit.
Apple could make their cards have a secondary adapter on the cards lower edge that is raised higher than the PCI-e standard allows in-between the PCI-e slot the outer edge of the card (the plug on the board sticks up higher ( 1-2" ) than std PCI-e socket) . That way standard PCI-e won't sit properly. the space on the board to hold the electronic to suppose the physical display output connectors isn't needed so can just remove a significant notch from that area the raised connector fits into. ( Similarly, could expand unified connector larger so card doesn't fit. ). The Apple could fit in the standard PCI-e slot (with the video display unconnected) but not vice versa. That drops the hocus pocus "requirement" for an adapter the standard cards.
Apple may very well sell 2 GPU BTO configurations, but in the compute GPU role the video display output just isn't needed. Fitting into a standard PCI-e slot gives all the functionality an Apple computer GPU would need. The variance on the slot isn't for PCI-e data (it isn't a new PCI-e slot). The variance on the slot is for video data output. That's a new standard Apple perhaps could propose. It won't be a PCI-e standard though. The PCI-e committee walked away from MXM cards. If it is not purely 100% PCI-e they won't adopt it. So what standard group is to take on PCI-e+DP .... probably nobody. This is only really a driving issue when have > 2 TBs sockets ( the 4-6 range the Mac Pro will likely have). Most of the workstation folks are provisioning just one. The mac Mini has more than one. That crowd isn't going to jump on board. Neither are the race to the bottom price GPU card folks. The laptop folks? Nope (card and edge too big).
Same for SSDs. Apple can use theirs for the boot SSD, but no good reason for a 2nd ( or 3rd or 4th) SSD socket not to be M.2.
But even, I don't see Apple allowing STD PCIe slots at all, beyond TB3 or to license the new PCIe form factor to VARs, Apple will lose thousand dollars on each mMP only in GPU upgrades, it wont be approved by Tim Cook neither Stock Holders.
While I don't see Apple providing two x4 physical slots like the older Mac Pros ( most of the stuff that folks pushed into those slots can be done either through M.2 or out through Thunderbolt). However, neither M2. nor Thunderbolt going to cover the x8-x16 PCI-e v3 bandwidth range. Apple is suppose to be making a Mac Pro that does not have higher bandwidth limitation problems. One x16 slot would give some. [ Yes there are those who will ask for 2, 3, 4 allocatable, x8+ slots but the numbers diminish as those numbers go up. ]
if the boot/default GPU cards only fit then Apple would get boot display upgrade business. So done. However, that wouldn't be a "change bottom line" business. There nothing material for the stockholders to see there. The tactical objective those would have would be to raise to volume of GPUs sold so Apple could hit breakeven on the cards more easily. What Apple needs is to allocate R&D resources so they can at least drop one new GPU every 8-18 months. Apple's
primary major failure has been playing Rip Van Winkle. Solve that problem and they will probably make more money selling the systems (not the upgrades). They can increase Mac Pro sales by simply doing something on a regular basis.
Strategically, lack of the internal PCI-e expansion means the very large fraction of the people in this subsegment more likely just won't buy the Mac Pro at all (if passed on iMac Pro can easily just well pass on Mac Pro if has same limitations). Loosing the $3,000+ Mac Pro sales ( $600+ profit) to goose some extra $100-200 profit on upgrade card is the tail wagging the dog. It is a complete bozo move. The compute GPU , audio / video capture , high end specialty cards , and even non-boot bulk storage (RAID cards) are all post boot. Trying to funnel all that through an Apple socket PCI-e sockets buys a whole lot of nothing.
Vice versa also though. If what folks primarily need is a $400-900 GPU upgrade forcing them into $3,000 purchase isn't going to work well over the long term. Even more so when detaching your system from the $3-10K other elements of their system that don't have to to with boot display functionality.
Once Apple drops standard configuration requirement that for dual GPUs there is x16 lanes of bandwidth just dangling out there. Assigning all of that to provisioning an absurd number of Thunderbolt ports is grossly silly. If taking largish cards inside then best fit is that standard PCI-e card for factor. A x16 socket can take x1 , x4, x8 , or x16 cards. There is a wide and varied collection of those cards customers have a sunk cost attachment too. Provide the slot and you get them.
The question is really about whether Apple allows 3rd party cards in at all. Allowing them means volume, cooling , and power requirement need to have lots of slack added to them. It would probably be bigger than what they'd want to put on a desktop. If Apple is still stuck on "it has to be a literal desktop" then I'd say pragmatically no slots. Apple has a desktop solution in the iMac Pro. I don't know why they'd need two at the high end. The Mac Pro really wasn't at literal desktop before. There is no "PCI-e socket form factor" to license because PCI-e (or its standard large card socket), in and of itself, really isn't the issue.
[doublepost=1520979210][/doublepost]
the imac pro is limited by it's slots a workstation can have room for more slots
The previous Mac Pro has the same 4 DIMM limitation (per CPU package) put on them. The Mac Pro 2009-2012 had them. That is is partially the consistency. The CPU tray couldn't hold dual rank ( 8 DIMMs per CPU package) in the dual CPU package configuration. So Apple just did 4 DIMMS / CPU whether had one or two CPU packages. [ had some secondary effect of pushing folks who wanted 8 DIMMs into buying two CPU packages so didn't hurt Apple's bottom line either. That is a much higher $/DIMM slot cost. ]
The iMac Pro doesn't have room. The Mac Pro if using Xeon W would not have a dual CPU package option to push folks into buying a second CPU package. Apple should put some gap between Mac Pro and iMac Pro and there is probably room in a Mac Pro to do it , so could go dual rank. And no good reason to avoid it. Once Intel (and AMD) split the workstation CPU package away from the multiple socket one that game of jacking up $/DIMM slot looses effectiveness.
I don't think Apple is a big fan of empty sockets/slots. That is just an invitation for people to open the box and stick something in them. The Mac Pro never really left the "open the box" camp.
Long term it isn't really a big limitation on the iMac Pro. The next Xeon W upgrade probably will support higher density RAM chips on the DIMMs. The iMac Pro limit will go up over the next year or so with mainly just CPU configuration upgrade. The Mac Pro would be a better option shorter term for those who needed big RAM disks ( or extreme large RAM working set) storage with a smaller budget allocation for RAM.
[doublepost=1520980837][/doublepost]
....
...which is something that Intel will have to sort out with PCI-SIG, VESA, NVIDIA and AMD anyway if they want to push TB3 technology into PC workstations, make TB3 the professional video connector of choice etc.
The PCI-SIG isn't going to touch anything that isn't 100% PCI issue. They don't even do the laptop MXM factor ( dropped). VESA ... likely same issue. It is a cross over system level issue.
Nvidia and AMD ... don't really care and as long as cyrptocurrency compute money rains out of the sky .... care even less. They really didn't get together all that much over MXM, why would they on larger cards?
TBv3 doesn't require directly coupling to the GPU card but it extremely much cleaner design to do so. For two or more TB controllers it gets even more messy.
What it does is reduce the issue from third parties needing to produce unique hardware to work with the mMP to the much simpler and cheaper issue of custom firmware and drivers. NB: NVIDIA are still providing MacOS drivers for their PCIe cards for the benefit of the Hackintosh and Cheesegrater communities.
The unique hardware isn't a barrier. If there were 80M mac pros that card vendors could make money on and it cost $10 extra to make the alternative card they would. Market size is a primary issue here. Not just in GPU cards, but all the standard PCI-e sockets. There are no Infinband cards for the Mac Pro. the market is small and the market for those who'd want to put those in the Mac Pro is even smaller.
The notion of "if you build it they will come" hasn't really proven itself in the Mac Pro market. Apple left lots of MP users stuck after 2013 and how many first party vendors showed up with a bootable GPU card to sell. That's the issue.
For small ecosystems the bundled GPUs in the standard configuration is likely to be the one cards that get to volume breakeven sales.
All that said, I'd have to go with:
(6) Partly sealed unit with minimal upgradeability (RAM and - if you're really lucky SSD) and your choice of external Thunderbolt 3 GPUs.
it really isn't only about GPUs . Throw out GPUs and look at the rest of the PCI-e card market. Find a reasonable solution and then weave back in GPU to that. ( that won't exclude that GPU solutions in any substantive way, but gets rid the myopic thinking that runs through these threads. ) Apple could view working on an Mac Pro update that way but they would be just be painting themselves into another, but very similar, corner.
Once the CPU and GPU cooling solutions are decoupled. And the GPU is running at a very high TDP.... the standard reference card design basically works. Apple could go to a more rational cooler ( that isn't restricted to double slot in width), but the MP 2009-2012 had a CPU tray (card)..... this is just about as high a TDP problem.
