Sorry, but no. Even Apple needs to conform to industry standards. Why would anyone make add-in cards for a proprietary platform with a limited market when they can just make a standard PCIe card that will work on a number of different machines? It's not about DIY, it's about producing a tool that enables people to do their work.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
[POLL] What are the Odds to have PCIe Slots in the new Mac Pro
- Thread starter Mago
- Start date
- Sort by reaction score
Sorry, but no. Even Apple needs to conform to industry standards. Why would anyone make add-in cards for a proprietary platform with a limited market when they can just make a standard PCIe card that will work on a number of different machines? It's not about DIY, it's about producing a tool that enables people to do their work.
hardware raid for 2 ports? no as it will be that fake raid BS. With 2 Xeons they have the pci-e for 4 m.2 ports and 4 TB3 buses + at least 2 X16 slots and dual 10-gig networking.
In the iMac pro it's rumoured the raid is essentially hardware based, with the T2 chip handling raid.
And the Mac Pro video card market is what? relatively small runs or large ones? It is relatively small. That's why there historically (last decade or two ) relatively few "Mac" video cards that have been brought to market. The default cards sold as Apple standard configurations are by far the largest number of cards sold. So not only a relatively small market but Apple is going to get most of the money in pool. The secondary cards are going to what is left over in a already small pool.
That's why Apple is probably going to "own" boot display solution. The number of Mac Pro's sold is going to at best plateau ( if not decrease from 2008-2010 levels ). The contribution of eGPUs probably would be an offset a discrete GPU decline even for the impact on the overall Mac market.
That isn't strictly true (there are more than one and the nuance is off). Three major points. First, the nuance was to not tie two largely asymmetric heat sources to the same heat sink. If the mid-power GPUs were in the same TDP range +/-15% as the CPU the system would have worked better. The issue was that GPU were tracking more along the lines of having double the TDP of CPUs that Apple was likely to use in a Mac Pro.
The applied that "reduce the coupling" lesson to the iMac Pro. There are two fans. There are two radiator loops ( one for each GPU and CPU) and don't tightly coupled with each other. Share the same exit vent and maybe some very mild heat bleed across the two loops (due to close proximity at vent), but very substantially decoupled.
The Mac Pro would likely allow them to decouple it even more. They don't really need to share the exact same exit vent that is nominally hidden behind the pedestal arm. Hiding the vent was the primary driving force coupling the exit vents on the iMac. Two fans probably means don't have a single cone shape wrapped around the single relatively large fan.
The Mac Pro 2013 GPU card sat on a daughter board in a socket. Once decouple and assign its own heatsink and cooling fan that really doesn't inhibit sticking with using a socket in any way.
Second, one of the other significant admissions was that the MP 2013 ( and likely the iMac Pro also )
"... In the interim, we know there are a number of customers who continue to buy our current Mac Pro. To be clear, our current Mac Pro has met the needs of some of our customers, and we know clearly not all of our customers. None of this is black and white, it’s a wide variety of customers. For some, it’s the kind of system they wanted; for others, it was not. ..."
https://techcrunch.com/2017/04/06/t...-john-ternus-on-the-state-of-apples-pro-macs/
It is clear from that they can't do another very highly overlapping system like the MP 2013. The higher "horsepower" iMac is probably good for most of the folks who were OK with the MP 2013, but the set of needs the folks left parked on older Mac Pros (and folks that left) are substantively different. A MP 2013 concept pushed into a rectangular box with two fans isn't really going to solve the market issue.
" .... and we want to architect it so that we can keep it fresh with regular improvements, and we’re committed to making it our highest-end, high throughput desktop system, designed for our demanding pro customers. ..."
Lacking regular improvements was a major problem. With some reasonable periodicity Apple needs to drop incremental improvements. An extremely effective way of doing that would be to drop a new GPU card into the system on a regular basis tweaking the standard configs. Similarly highest throughput is more than hampered if they block x16 lanes off the CPU being being used. Apple also outlined that dual GPU as a standard configuration wasn't a good fit ( perhaps buried in that 'two mid power GPU' point). So what high throughput task is assigned to the lanes that Compute GPU was sitting on? And why is that decoupled for the wide diversity of users? Nothing particularly indicates that Apple doesn't get those points. Some people in the Mac Pro market need a Compute GPU and others new some other high throughput solution. If everyone needed Compute GPU then dual GPUs would still be on the "standard configuration" target. They aren't.
In the MP 2013 the pins on the Compute GPUs DisplayPort path were dead. It was just symmetrical consistency to use the same adaptor on both and two were always in use (so just buy twice as many of the same custom adapter). If going to enable a high throughput Compute GPU that Apple may or may not sell then a standard PCI-e slot works far more better for a much wider range of solutions.
I think Apple understands all to well that a huge fraction of these "other" Mac Pro customers are going to apply a 'form over function' argument of PCI-e slot to to the boot display GPU. ( the standard slot doesn't ingrate cleanly with Thunderbolt at scale. It doesn't meet the function. ). So in addition to not talking about details of future products are also going to avoid "drama" of standard PCI-e slots should be the only solution you use discussion.
Again this somewhat boils down to not overdoing coupling. Decouple the "Compute GPU" socket from the "Display GPU" socket since they have substantively different functions. Unnecessary coupling paints you into corners that can be difficult. That doesn't mean that Apple's objective is to be a biggest container of the widest variety of parts. High Throughput is an objective. So that decoupled probably had high priority. That someone has 6-8 2.5" drives they want to stuff into a Mac Pro container box is probably not a priority.
[doublepost=1521048041][/doublepost]
All modern reasonable SSD controllers "raid" the NAND Flash chips. If your SSD reads and writes at roughly the same speeds, then it is using are variant of RAID.
There is no SSD controller on the two "cards" attached to the T2. It is pragmatically an SSD split into two daughter boards.
[doublepost=1521050139][/doublepost]
Going from Intel W solutions to Xeon SP solutions is moving to more affordable Xeon solutions. In most cases it is moving to a higher price point not a lower one.
You can toss performance out the window and find a 8 core SP that matches core count with an 8 core W.
$417 Base clock 1.8GHz turbo 3.0GHz
https://ark.intel.com/products/123544/Intel-Xeon-Silver-4108-Processor-11M-Cache-1_80-GHz
$1,113 Base clock 3.7GHz tubro 4.5GHz
https://ark.intel.com/products/126707/Intel-Xeon-W-2145-Processor-11M-Cache-3_70-GHz
but if try to match base clock
$2925 Base clock 3.5GHz turbo 4.2GHz
https://ark.intel.com/products/124943/Intel-Xeon-Gold-6144-Processor-24_75M-Cache-3_50-GHz
That's is a > $1,000 increase in cost. That isn't the bargain discount path (and still incrementally slower).
You can buy 16 very substantively slower cores with Xeon SP silver products, but you are not going to get the same performance on marginally to low threaded apps. Frankly some leading edge MBP will probably beat the crap out of it too.
$500 Base 2.1GHz
https://ark.intel.com/products/123547/Intel-Xeon-Silver-4110-Processor-11M-Cache-2_10-GHz
so 2 * $500 = $1,000 bring into rought parity with 8 core W model.
16 * 2.1 33.6 hz*core metric
8 * 3.7 29.6 hz*core metric ( 12% decrease )
the above is base, but turbo ( 1 to 2 , 1-2 cores )
1-2 * 3 6 hz*core metric
1-2 * 4.5 9 hz*core metric ( 50% increase )
This yes. The dual CPU packages to get to 8 DIMM slots basically means paying higher $/DIMM slot unless completely fart away performance.
Probably not going to happen. Thunderbolt and/or SSDs will soak up some of those "lower bandwidth slots" since only have a smaller budget.
One of those slots slot suggest probably is going to be assigned to the display GPU.
Again the presumption of a huge glut of lanes due to dual CPU packages runs up against the reality of dual CPU pricing. 2 controllers ( 4 ports ) probably consumed your "lower bandwidth slots".
As noted before the T2 aren't two SSDs it is one SSD split into two logic boards. Those are custom, it is just the internals of an SSD exposed a bit more than usual.
However, the Mac Pro should have some M.2/NVMe sockets built in. More than one storage drive would be highly useful.
You might get 1-2 drive sleds but somewhat random SATA ports dangling on the end of cables probably not happening. The number of SATA sleds is probably down from previous systems. The 4 drive sleds of old were primarily there to do RAID. As I outlined if you have a SSD you have "RAID".
Wasn't particularly an option on previous Mac Pros. Kw range also probably not. The 2010 versions were in the 800-900W which is relatively high and was standard.
Last edited:
Right - most SSDs consist of a bunch of flash memory, and a controller that upstream provides a SATA/SAS/IDE/NVMe interface, and downstream stripes the flash memory into a pool of visible memory and a pool of hidden (over-provisioned) memory for the garbage collection.
The iMP does not have any SSD drives in the traditional sense.
It has the T2 chip, and a couple of proprietary daughtercards. The T2 chip's storage layer most likely appears to the OS as a standard NVMe device. Behind that, the T2 chip manages the flash memory on the two daughtercards.
At this point we don't know much about the interface between the T2 chip and the daughtercards. It's possible that the daughtercards are purely memory, and the T2 handles all of the memory allocation and garbage collection on-chip. It's also possible that the daughtercards are more-or-less complete NVMe drives, so that the T2 does the striping and encryption - but the daughtercards handle allocation and garbage collection.
The Xeon-W supports 8 DIMMs out of the box (four channels, two DIMMs per channel). Restricting the iMP to 4 DIMMs was an Apple design decision. A single socket mMP with Xeon-W could have 8 DIMMs.
The Xeon SP support 12 DIMMs out of the box (six channels, two DIMMs per channel). A pair of Xeon SP support 24 DIMMs.
The latter seems impractical. If there no chip big enough to be a controller then it isn't a full drive. Picture from the iFixit teardown
Step 8 https://www.ifixit.com/Teardown/iMac+Pro+Teardown/101807
The big red highlighted chips are the NAND chips. The notion that is is a full fledged SSD hinges on that very small orange highlighted chip being the SSD controller. There is nothing else on the card. iiFixit tags it as a Apple chips. Somehow there is one or two a ARM cores and NAND management circuits in there? Most SSD controllers are closer to the size of a NAND chip (in red) than that orange doo-dad. that looks more like a DIMM register. Something pawn off writing a buffer or two out onto the NANDs.
'
Thanks for the iFixit link!
One question now is whether the ARM CPU cores in the T2 chip are managing the SSD daughtercards - or if there is "ASIC-like" dedicated silicon in the T2 for the SSD controller. (Or a mix - the allocation/striping/garbage collection is in an "ASIC-like" processor, and the ARM cores handle the AES functions.)
T2 is a system on a chip that includes an SSD controller.
I don't know whether the RAID aspect is handled by dedicated hardware or the ARM chip. It is a faster chip than you'd typically see used for this sort of thing.
I'm still personally interested if T2 includes the graphics package, but my research so far implies it does have a GPU as well.
[doublepost=1521072556][/doublepost]
Right. This is where things get downright silly. They're going to allow every other Mac to work with an off-the-shelf-at-Best-Buy PCIe GPU externally, but then block it as an internal option on the Mac Pro?
It doesn't make sense.
If anything stops it, it will be technical issues and not religious issues.
The T2 is a system on a chip that controls an SSD (actually, that takes a couple of flash daughtercards and emulates an SSD for the host OS).
That's a very different statement.
And my comment was directed towards whether the T2 die has "ASIC-like" silicon to manage the daughtercards - or whether the ARM cores are doing all of the work.
The T2 does more than just SSD work. Sysetm power management. Secure enclave . there are a couple of things besides the SSD controllers subsystem that needs to do a little bit of "smart"/"brainy" stuff. It would make sense for the more brainy stuff to be done on a shared ARM core(s). For the SSD brainy stuff is things like wear leveling ( do I reuse this block or shuffle the to another one ).
Very high chance that the encryption ( AES) is almost all fix function logic ( ASIC). Compression also. the ecc/correction codes perhaps as well. This chip can likely compress and encrypt the data going to/from the storage at wire speed. With fixed function logic that isn't hard. It also isnt' particularly unique to the T2 (or T1 ). Basically Apple is likely borrowing the a large fraction of the SSD storage subsystem they have in the iPad Pro/iPhones. In read speeds Apple has been in this zone. They need to boost writes but there are more NAND chips/dies to spread the load here ( more RAID).
That may be background on the funky thing with two storage daughter boards as the iPhone/iPads are only use to dealing with 1-2 NAND chips. Put a "buffer register" between them and could expand to 4 on each board then the "brainy" part interleaves between the two boards. At its heart still a iOS implementation but using them in more parallel boost the speed. The 70M sold per year iOS devices sold each year basically have the R&D costs covered to the basic subsystem. T2 SSD controller is just putting a NVMe PCI-e interface on top and the "brainy" bits to handle boosting write speeds and better wear management.
[ the other reason the NAND is detachable is that it easier to retire those system while shredding the storage. I suspect a number of more secure org weren't happy with soldering NAND to the motherboard. ]
A watch S3 ARM cores ( for low power) and the fixed storage logic from the iPad Pro and latest iPhones.
May be an even smaller power management core that can sleep at even lower power. (motion and some phone/watch specific stuff possibly stripped off)
[doublepost=1521078767][/doublepost]
The T1 has a GPU to drive the touch bar. Even though the iMac Pro doesn't have a touch bar it doesn't make much sense to come up with different Tx versions for desktop versus laptops. Apple only sells 15-20M Macs a year. that is a relatively low run rate. [current macrumors front page story pegs Apple watches about 40M ). So probably going to ramp up to same chip is the laptops with touch bars too.
if detaching a screen puts the GPU almost entirely to sleep not really loosing much on power loss. If there is some smart math to be done ( or camera processing ) than can use it as embarrassingly parallel compute engine.
Can eGPU's handle a boot screen (handle secure unlocking of the system)? if not then I don't see the disconnect. Even if they do handle it, is it being handled by some redirect through the embedded GPU? Again in that case I don't see the disconnect.
There are two major factors. Operating in Apple secure boot environment. The second is likely integration with Thunderbolt. The eGPUs don't really have to integrate with TB. It runs over TB but TB should be opaque to the eGPU. So not sure that is a integration solution.
If there are a wider set of "happen to mostly work" GPUs I don't think Apple is going to be mad/upset ( as long as the number of support calls are low). However, I don't think Apple is going to count them as a viable, complete, certified system solution.
As the secondary x16 slot yeah it wouldn't make sense to exclude them. That isn't the default primary function though.
Last edited:
P.S. did some digging around on some other tear downs. Step 11 on the MBP 13" function key model has the detachable SSD.
https://www.ifixit.com/Teardown/iMac+Pro+Teardown/101807
that card has an SSD controller and an another slightly older, highlighted in orange, Apple doo-dad like the T2's cards. More doubt that the "Apple 338S00285" ( or Apple 338S00227 ) are a primary SSD controller. The other noticeable thing too is that without the RAM and the controller on board the whole thing is much cleaner looking (and decent chance easier to make). Would be curious to see what the 1TB modules look like (if a tradeoff was to pack more dies (in bigger packages ) NAND on the card with the support electronics.
Perhaps a bit telling that their SSD controller is not a RAM less controller (presuming Apple's stuff, not a design they licensed a custom firmware for). Yet another indicator. The RAM is probably attached to the T2.
I'm actually not sure. The EFI's Apple ships have some rough edges around eGPU support (like breaking it under Boot Camp) so there have to be EFI updates anyway. On laptops the EFI firmwares are generally hardcoded to look for the internal GPUs. That's solvable on a Mac Pro.
So ¯\_(ツ)_/¯.
Nope, it's direct only right now. GPU tunneling would be a very useful feature for using the eGPU with the internal display, but right now that's not how it works.
Right, which again goes back to, if it's not going to happen, it's going to be an engineering side problem.
[doublepost=1521134619][/doublepost]
Right, most people don't know that the A9 already has a full blown desktop grade SSD controller (thought to be NVMe) because of the iPad Pro.
(It looks like online people are pretty sure the A9 does have NVMe, so it's probably more than "thought to be.")
Most likely is it NVMe. The SSD controller from the iPhone/iPads just needs t agumented with a "front end" that 'speaks' PCI-e v3 and NVMe protocol. It takes bits in those formats and tranforms them to the protocol used inside the Apple SoC uses to communicate to the
Most SSD cards/devices look like this
----SATA/PCI-e---> [ controller ] ----> [ NAND chips]
if the controller is split
--> [ outside protocol ] -- inside protocol ---- > [ SSD "smarts" ] --> [ rest of SSD ]
The "smarts" doesn't really care where the chunk of data comes from. It just read and write what is requested. This is why many controllers have two small core to handle the administrative overhead of each halves. The hardware elements of SSD can be busy doing SSD stuff and the other core is busy handling r/w requests from the outside. It is two different major tasks. The cypto stuff could present as a different "address" on the PCI-e bus (or be provisioned on a separate PCI-e lane.)
The T2 probably has multiple hooks back into the rest of the Mac system. The low level system/power management have have to GPIO hooks. The hook to camera/display are appropriately distinct as well.
The Intel CPU (and/or I/O Platform Controller Hum , PCH, ) are built expecting a NVMe SSD. This easiest integration with the is to just provide one.
And therein lies the (first) rub: the more that the video subsystem deviates from the PC norm, the more costly it is going to be ... not only for 3rd Parties to provide as alternatives, but also for Apple to provide periodic updates. Plus there's still the question of if Apple will be willing to sell these "better BTO parts" to existing mMP customers.
Well, the first part of this is the same issue: will an existing mMP customer be able to buy improvements to plug into their existing hardware? Its a simple yes/no question ... although the iMac Pro's "RAM Upgrade" bit does make it sound like it will be very tightly controlled by Apple (BTW, has anyone yet noticed if/where Apple actually is providing prices for iMac Pro RAM upgrades?).
Similarly, it does a customer no {bleeping} good to have the OEM be the sole source of upgrades if/when said OEM then never delivers on their promise of having regular improvements. To a great degree, this is sounding more like Apple saying "what's good for Apple is for having an architecture that's easy for Apple to make incremental hardware improvements to keep the model appearing more fresh." But of course, we've all covered this point before, particularly if we start to remember that there will be an Apple beancounter who grimaces and wrings their hands at the prospects of increased QA costs for testing (N+1) 3rd party graphics cards that aren't being sold (profited on) by Apple.
This is a good observation & point, although I'm afraid that the Apple paradigm is going to remain the attitude that a single does-everything cable is more important -- to Apple -- than truly enabling the degree of configuration flexibility that's valued by these Mac Pro end users. FWIW, I do personally wonder just how much of the potential 'problem' perception goes away if the video-out port just isn't a MDP/TB2 or TB3/3.1 in physical form.
Well, the good news is that as has been demonstrated on MBP's and M.2, SSD interfaces which are substantially faster than SATA-III are a mature technology today, so these can clearly take up much of the slack of where multiple 3.5" bays RAID'ed together were the prior solution.
However, it still would certainly be nice to have a bay for a ~6TB 3.5" SATA so as to have one's basic data backup system (via Time Machine or other software) onboard, and not sucking up network bandwidth, or adding an external box & cables. Better even to have two such bays, to permit an alternating A/B backup solution.
And finally (& FWIW), lets not neglect other supporting elements, such as a keyboard which doesn't suck as bad as the MBP ones....
-hh
SATA is one of the few things Apple was right to ditch. The bandwidth and latency is well covered by Thunderbolt, and in fact the 3rd parties have already built a semi-healthy market scaling from single drive enclosures to rack mounted RAIDs. Also there are realistic professional use cases that no longer need local storage, so the internal or DAS only acts as the boot volume and scratch/cache, where it is trending towards direct PCIe for obvious benefits.
Using a DAS/NAS for workload Data/Backup has become not just clever, also convenient, saves time, money, etc.
It wastes space on the motherboard and back panel. Almost anything else is a much better use for the space.
but if you have a big tower it should be there and if not at least 2 real M.2 slots
Big tower is doubtful as that doesn't fit Apple's design language. The iMP doesn't have m.2 but another, worse non-PCIe connector.
Management of 'available space' and packaging efficiency is a legitimate concern for a mobile device ...
... but who said that a (modular) "Pro" desktop system has to be a highly compact mobile device?
No one.
Considering how packaging density generally runs directly counter to good thermal management design principles...
(even before we call how the prior 'Pro' desktop was a thermal management dead end because of how it was unnecessarily constrained because of the foolish/egotistical design aesthetic _desire_ for it to be "small")
...what you're doing is unnecessarily adding the burden of an additional constraint to your product design, which makes it (a) harder to design; (b) harder to produce; (c) harder to maintain; (d) less capable at running at full throttle; etc.
And note also that:
Factor (a) = more expensive @ supplier
Factor (b) = more expensive @ supplier
Factor (c) = more expensive @ customer (lifecycle costs)
Factor (d) = more expensive @ customer (lifecycle costs)
Not no one. If Apple is trying to literally put the Mac Pro on the desktop most users have system footprint constraints for their desktops ( largely because there is other stuff they need on their desks besides just the base computer system). The Mac Mini has a reduced footprint because of this need. All-in-One place the system behind the largish monitor to minimiize (effectively reduce to zero) the actual physical footprint on the desk. To say that no one cares about the size of stuff on their desk is an exaggeration. More than a few do; on one simply just isn't true at all.
Another way of getting to zero print on the desk is to move the whole system to desk-side ( or under desk or just plain remote ). Monitor , keyboard , and mouse (and perhaps a dock for front facing ports) are on the desk and the whole system is moved to a slightly more remote location ( side/under desk). In that context, "small as possible" isn't an issue. Apple probably isn't going to do large as possible, but there are fewer constraints.
Random off-the-shelf component with a mishmash of thermal design philosophies also run counter to good thermal design principles. The notion of just be an overly large container that can deal with arbitrary thermal designs is something that Apple is probably going to be resistant to.
The MP 2013 was design right up the the thermal tolerances. That was a bad move. If Apple had put just a reasonable bit of slack into the overall system design there would have been less drama with some thermal failures. (e.g., if target if 430W then do 500W and have a slack buffer of capacity. )
simplification in avoiding random, nonaligned solutions is not harder to design or harder to produce. nor maintain , nor full throttle. It just keeps out stuff that doesn't fit well which random user wants to jam into the box anyway because they can.
[doublepost=1522175378][/doublepost]
That the same kind of baseline myopic mantra that keeps BIOS support alive. Gotta design for the lowest common denominator in the race to the bottom market. Intel is the only one who announced dropping BIOS and that's years from now
https://www.anandtech.com/show/12068/intel-to-remove-bios-support-from-uefi-by-2020
[ EFI started out in 2005-6 . UEFI started in around 2007 and ... 11 years later still have transitioned off of BIOS. ]
So harping on Apple because they are keeping their EFI from when they jumped on board with both feet early and quickly and the rest of the overall dragging their feet as slow as possible with BIOS. If everyone conformed 100% with what everyone else was doing you'd get almost not progress.
If it is just a slot pull then sooner or later the parts would flow. Retired systems would be a source of parts. If the prices are higher than the norm most of the folks who are price anchored to the market aren't going to buy anyway until they drop to or below those norms. Well folks something yelp about the bleeding edge cards they can't get, much of what is going on with the Mac Pro currently is buying newish/used cards for substantially older Mac Pros. That is a substantive level of the noisy grumbling going on.
That's said if Apple has their own slot format that is not as big of an issue if they chose something and consistantly stick with it across 10-15 years of upgrades. Once there is a long term flow there will be a market (if only in the newer cards for newer systems are being placed in older ones. )
If they put a single standard x16 slot in that would be answer whose answer is self evident. Fighting for control over the primary display GPU slot and a second slot should be separated into two different issues. Folks who are control to continue to grumble that they couldn't also get to the primary slot, largely just want to grumble. It will also be something until Apple has backpedaled into the lowest common denominator in the race to the bottom market.
If Apple delivers then what is the issue? The complaint starts off as Apple not delivering being the issue. Well, Apple's statement you quoted basically alludes to them saying that isn't good for the Mac Pro. So if they fail the whole thing is pretty much going to fail .
If baseline mainstream market has shown solving the core technical design issue. PCI SIG want to be PCI-e pure. The general AIO card market is oblivious to Thunderbolt and point to Rube Goldberg kludges to get around the problem. MXM cards are really somewhat of pseudo standard that on a subset of solutions are produced for. etc. etc.
If the mainstream market was actually solving the issue with a solid solution standard and Apple was not falling in line then sure .. .that is a pain. This reality is that they aren't trying at all. Apple isn't a "monkey see , monkey do" company which is going to simple just pile into the same straight jackets . That is the fast track to becoming a dead, formerly large, leading, tech company.
They didn't sell in eye popping numbers before. It isn't profit lust after what other vendors are getting.
To the vast majority of modern Macs they are important. So that should make it important to Apple since the Mac ecosystem is a Fortune 500 sized business for them.
Flexibility isn't really the issue. It really is cheap and price anchoring that is the real issue. All too simple solution is to make the secondary slot a standard x16 . That's flexible. Done. More than one would be incrementally flexible but one covers a quite sizable chunk of the market. The other issue is what apple alluded to. Produce regular updates. If that turns into "I refuse to buy those from Apple" then that really isn't flexibility issue at all.
Really? You haven't been in forums about USB Type-C cables where folks are yelping about why there are 2-3 different kinds of cables. Some that do mainly just power (and only get USB 2 ) , some that can do USB , Power , and TBv3 , and others that do TBv3 but not power.
There are a huge segment of users that have been highly conditioned into a mindset of "if it fits it should work".
Internal to the system backups seem like an odd duck. Most of the single points of failure within the system being backed up are exactly the same single points of failure of the backup itself. ( same power supply, same controller , .... )
It is better than none but some backup needs to be decoupled.