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.
MP All Models What a new mac pro 2022 needs to be?
- Thread starter innerproduct
- Start date
- Sort by reaction score
In April 2017, the one explicit use of "Modular" was in the exact same fashion it is used on the Studio. The display is modular.
" ...
As part of doing a new Mac Pro — it is, by definition, a modular system — we will be doing a pro display as well. ... We think it’s really important to create something great for our pro customers who want a Mac Pro modular system, and that’ll take longer than this year to do...
.... I think, as you talk about the pro user, the fact that our user base is split over notebooks, all-in-one desktops and modular desktops is important. ...
.... who use iMacs and who use Mac Pros, who use modular systems as well as all-in-one systems, who use the pro software we make. ... "
Transcript: Phil Schiller, Craig Federighi and John Ternus on the state of Apple's pro Macs | TechCrunch
You have already read the news. But we thought we would also use this opportunity to share a transcript of the interview with Phil Schiller, Senior Vice
techcrunch.com
Modular was consistently juxtaposed against the all-in-one iMac ( and iMac Pro which is heavily being hinted at in a couple of places in this transcript. This session wasn't solely about the Mac Pro... it is also basically an intro preparation for the iMac Pro (which was at that time a higher priority). )
Several aspects of the Mac Pro 2019 modularity that were pragmatically forced on Apple was that the Xeon W 3000 series was iGPU less. If pick that as a CPU then have to go to discrete GPUs. Apple had no choice. In 2020+ they do have choices because Intel is out of the primary SoC design choice position.
Same issue with RAM DIMMs. Basically an Intel driven design choice that Apple had to provision for.
As for GPUs...
" ...
We designed a system that we thought with the kind of GPUs that at the time we thought we needed, and that we thought we could well serve with a two GPU architecture… ..... But workloads didn’t materialize to fit that as broadly as we hoped.
Being able to put larger single GPUs required a different system architecture and more thermal capacity than that system was designed to accommodate. So it became fairly difficult to adjust. ..."
So they got on the large single GPU quest. Careful for what you wish for. Four Max sized dies 'glued together' on a single SoC is probably going to be one of the highest die space allocations for a single GPU that you can get in next year or so. Most of the Max die is the GPU and associated Memory and Video de/encode and display controllers. Four of them will be a large single GPU.
Getting apps to fully leverage the double and quad "Max like" GPU core clusters will probably 'materialize' better than the "two GPU , OpenCL workloads" did for the Mac Pro 2013 set up. Apple has a much more solid software foundation laid out at this point for that to happen on. And "unified memory" makes it easier to scale up once in the guidelines that Apple has laid down for the last several years. 12-18 months Apple allows 3rd party developers to optimize for their mega sized , single GPU then updated apps will probably turn in pretty decent performance.
Apple also said
"... We have a team working hard on it right now, 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. ... "
Note that it is "we can keep it fresh". It isn't so the end users can keep it fresh by regularly ordering new parts from Frys/Newegg/Amazon/Microcenter/etc. It is so Apple can roll out updated product. ( and yes when they do that via modular means there is secondary fall out effect of the 3rd party stores being alternative options. But Apple isn't directly talking about making as many alternative purchase channels as possible. )
High Throughput? The memory subsystem on the Ultra is pretty high. Double Ultra would be probably be twice as high. Decoding double digit number of 8K RAW video streams? That is high throughput.
What is missing is PCI-e v4 (or v5) general I/O in 'bulk'. Hanging that off of an augment to UltraFusion wouldn't be took hard. ( put two x16 PCI-e v4 controllers and would have more than decent bandwidth/throughput given the GPU is covered internally and not a necessarily a huge consumer. )
The other thing that has changed over last 5 years is that there is a bigger pool of software that can leverage "Compute GPUs". For "embarrassingly parallel" workloads chopping workloads up into more chunks and spreading them out over multiple "compute accelerators" works well. Fabrication and packaging constraints limit just how big can make a single , large GPU... and two-three largish ones can spread more load than a singles large one can. Apple may not want to let "display GPU" drivers back into the mix, but "compute accelerator" drivers will be a gap for highest end workloads.
If the SoC "card" has the RAM , SSD drive , probably the TB sockets (as the primary video out ports ) .... what exactly is substantively left on the rest of the board?
Replace cost for that or replacement cost for a motherboard is going to be significantly different how?
[ compared to Mac Pro 2019 board can dump the DIMMs slots, relatively complex DisplayPort routing, dual feed PCI-e switch. most of that got subsumed into the SoC package. ]
Neither macOS nor their Apple GPU drivers support that "Multiple SoC" notion at all.
Apple has at every opportunity explicitly cheerleaded the notion that "Unified Memory" is 'insanely great' and a key lynchpin to their approach. Pretty unlikely they are going to throw that out the window at this point for a small single digit percentage, practically "hobby" product.
In April 2017, the one explicit use of "Modular" was in the exact same fashion it is used on the Studio. The display is modular.
" ...
As part of doing a new Mac Pro — it is, by definition, a modular system — we will be doing a pro display as well. ... We think it’s really important to create something great for our pro customers who want a Mac Pro modular system, and that’ll take longer than this year to do...
.... I think, as you talk about the pro user, the fact that our user base is split over notebooks, all-in-one desktops and modular desktops is important. ...
.... who use iMacs and who use Mac Pros, who use modular systems as well as all-in-one systems, who use the pro software we make. ... "
Transcript: Phil Schiller, Craig Federighi and John Ternus on the state of Apple's pro Macs | TechCrunch
You have already read the news. But we thought we would also use this opportunity to share a transcript of the interview with Phil Schiller, Senior Vice
Modular was consistently juxtaposed against the all-in-one iMac ( and iMac Pro which is heavily being hinted at in a couple of places in this transcript. This session wasn't solely about the Mac Pro... it is also basically an intro preparation for the iMac Pro (which was at that time a higher priority). )
Several aspects of the Mac Pro 2019 modularity that were pragmatically forced on Apple was that the Xeon W 3000 series was iGPU less. If pick that as a CPU then have to go to discrete GPUs. Apple had no choice. In 2020+ they do have choices because Intel is out of the primary SoC design choice position.
Same issue with RAM DIMMs. Basically an Intel driven design choice that Apple had to provision for.
As for GPUs...
" ...
We designed a system that we thought with the kind of GPUs that at the time we thought we needed, and that we thought we could well serve with a two GPU architecture… ..... But workloads didn’t materialize to fit that as broadly as we hoped.
Being able to put larger single GPUs required a different system architecture and more thermal capacity than that system was designed to accommodate. So it became fairly difficult to adjust. ..."
So they got on the large single GPU quest. Careful for what you wish for. Four Max sized dies 'glued together' on a single SoC is probably going to be one of the highest die space allocations for a single GPU that you can get in next year or so. Most of the Max die is the GPU and associated Memory and Video de/encode and display controllers. Four of them will be a large single GPU.
Getting apps to fully leverage the double and quad "Max like" GPU core clusters will probably 'materialize' better than the "two GPU , OpenCL workloads" did for the Mac Pro 2013 set up. Apple has a much more solid software foundation laid out at this point for that to happen on. And "unified memory" makes it easier to scale up once in the guidelines that Apple has laid down for the last several years. 12-18 months Apple allows 3rd party developers to optimize for their mega sized , single GPU then updated apps will probably turn in pretty decent performance.
Apple also said
"... We have a team working hard on it right now, 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. ... "
Note that it is "we can keep it fresh". It isn't so the end users can keep it fresh by regularly ordering new parts from Frys/Newegg/Amazon/Microcenter/etc. It is so Apple can roll out updated product. ( and yes when they do that via modular means there is secondary fall out effect of the 3rd party stores being alternative options. But Apple isn't directly talking about making as many alternative purchase channels as possible. )
High Throughput? The memory subsystem on the Ultra is pretty high. Double Ultra would be probably be twice as high. Decoding double digit number of 8K RAW video streams? That is high throughput.
What is missing is PCI-e v4 (or v5) general I/O in 'bulk'. Hanging that off of an augment to UltraFusion wouldn't be took hard. ( put two x16 PCI-e v4 controllers and would have more than decent bandwidth/throughput given the GPU is covered internally and not a necessarily a huge consumer. )
The other thing that has changed over last 5 years is that there is a bigger pool of software that can leverage "Compute GPUs". For "embarrassingly parallel" workloads chopping workloads up into more chunks and spreading them out over multiple "compute accelerators" works well. Fabrication and packaging constraints limit just how big can make a single , large GPU... and two-three largish ones can spread more load than a singles large one can. Apple may not want to let "display GPU" drivers back into the mix, but "compute accelerator" drivers will be a gap for highest end workloads.
If the SoC "card" has the RAM , SSD drive , probably the TB sockets (as the primary video out ports ) .... what exactly is substantively left on the rest of the board?
Replace cost for that or replacement cost for a motherboard is going to be significantly different how?
[ compared to Mac Pro 2019 board can dump the DIMMs slots, relatively complex DisplayPort routing, dual feed PCI-e switch. most of that got subsumed into the SoC package. ]
Neither macOS nor their Apple GPU drivers support that "Multiple SoC" notion at all.
Apple has at every opportunity explicitly cheerleaded the notion that "Unified Memory" is 'insanely great' and a key lynchpin to their approach. Pretty unlikely they are going to throw that out the window at this point for a small single digit percentage, practically "hobby" product.
wrong and out of context bad quotes. If being headless made it modular he then Mac mini was modular. So was the failed trashcan. Far far off reality.
Mac Pro needs to nothing like the 2013 mac pro. Done.
If they solder the memory it better be HBM3 memory, then I won't mind and it should have slots. Apple sgould also make dGPUs.
If they solder the memory it better be HBM3 memory, then I won't mind and it should have slots. Apple sgould also make dGPUs.
i'm no pro user, but IMO the next mac pro needs to be a "box with slots". exactly to what degree is hard to tell. slotted cpu's, discrete gpu's, and upgradeable RAM is unrealistic, there's no way they will do that. my guess is that the slots will be strictly for i/o and storage, in some kind of shortened mpx style module/heatsink enclosure. i highly doubt it will be PCIE slots, apple will probably engineer yet another proprietary connector, which sucks but is expected. maybe they will make the connector licensable or something so that third parties can make cards, but that's pretty unlikely. these slots will probably be the only form of "upgrade", which is a pretty loose term in the age of AS.
going on a slight tangent, i don't consider the 2013 trashcan to be a "real mac pro", it's basically the mac studio before the mac studio - a non-expandable compact workstation, which is a totally different product class than a box with slots. when i think of "mac pro", i think of the old 2010 style and 2019 style - a box with slots, fully upgradeable and expandable using standard components and connectors. hopefully the new AS mac pro at least tries to keep in line with that philosophy. honestly i just wish the 2019 mac pro was bit cheaper, since it's pretty much the closest thing to a "standard pc" apple has ever made in recent history, and just like the 2010 people will still be using it 10 years later. no doubt in ~5 years it will become much cheaper secondhand as all the real pros either switch to windows/linux or their pro apps work on AS, and especially once data centers start liquidating their xeons for a few hundred a piece.
hopefully the rumor of an intel mac pro refresh is true, and they actually do it justice with the latest xeon lineup, pcie 5.0, and all the other modern features that were lacking from the 2019... and this time make it space grey
going on a slight tangent, i don't consider the 2013 trashcan to be a "real mac pro", it's basically the mac studio before the mac studio - a non-expandable compact workstation, which is a totally different product class than a box with slots. when i think of "mac pro", i think of the old 2010 style and 2019 style - a box with slots, fully upgradeable and expandable using standard components and connectors. hopefully the new AS mac pro at least tries to keep in line with that philosophy. honestly i just wish the 2019 mac pro was bit cheaper, since it's pretty much the closest thing to a "standard pc" apple has ever made in recent history, and just like the 2010 people will still be using it 10 years later. no doubt in ~5 years it will become much cheaper secondhand as all the real pros either switch to windows/linux or their pro apps work on AS, and especially once data centers start liquidating their xeons for a few hundred a piece.
hopefully the rumor of an intel mac pro refresh is true, and they actually do it justice with the latest xeon lineup, pcie 5.0, and all the other modern features that were lacking from the 2019... and this time make it space grey
Zero need for MPX slots if there are no discrete GPUs...
The extra slot on the MPX was for massive power delivery to the GPU (because Apple did not want cables snaking about in their Cheesegrater 2.0 chassis) and for Thunderbolt connectivity...
ASi Mac Pro will (most likely) have four regular PCIe slots for all your audio/video I/O, networking, and NVMe PCIe SSD RAID card needs...
But MPX slots...? Ancient history...
The extra slot on the MPX was for massive power delivery to the GPU (because Apple did not want cables snaking about in their Cheesegrater 2.0 chassis) and for Thunderbolt connectivity...
ASi Mac Pro will (most likely) have four regular PCIe slots for all your audio/video I/O, networking, and NVMe PCIe SSD RAID card needs...
But MPX slots...? Ancient history...
There is about zero upside is designing some entirely proprietary competitor to PCI-e standard slots. The MPX bay design is exactly illustrative of that in the Mac Pro 2019. Nothing has particularly changed over the last two years with Apple Silicon to change that.
Slots 1 and 3 in the MP 2019 have two connectors. One is regular standard x16 PCI-e v3. The other is the MPX connect that does the augmented stuff the MPX modules need. There were no 3rd party cards except for a limited (likely misadventure) storage offering in the form of the Promise R4i RAID MPX module. So if Apple went even more proprietary there would more likely be less than one 3rd party card.
It is bad enough that there is no UEFI in these new system so that many of these cards won't be effective at boot (unless there are general Apple written utility driver can leverage ). So long term need Apple new , non-kernel extension driver solutions. So the initialize and software stack are substantively different. If though substantively different electrical on top of that... most card vendors just aren't even going to even evaluate it, let alone implement.
Apple already supports some PCI-e cards in external Thunderbolt PCI-e enclosures already for existing M1 systems. So also about zero need to re-invent the wheel here since already have that to do. Mac Pro would be extremely similar with the slots just on the inside of the box using the same driver stack. There would be many cards that could be used across the whole Mac ecosystem; not just the Mac Pro.
The Mac Pro could probably be provisioned to handle bandwidth the external PCI-e enclosures couldn't. For example Sonnet just released this card
Sonnet Builds 8-Port 10Gbps USB-C PCIe Adapter
Sonnet launches the Allegro Pro USB-C 8-port PCIe 3.2 Gen 2 card.
Sitting in a x8 PCI-e v3 (or better) capable slot inside a Mac Pro could get the full bandwidth out of the aggregate ports on that card. Sitting in a Thunderbolt external PCI-e enclosure you cannot, but can get a useful chunk of it. That card has some utility for the Mac Pro and the entire rest of the M1 line up which has an external enclosure attached. That is a plain higher overall Mac ecosystem value add. The card vendor has a larger Mac market to sell into (bigger user base to amortize customer driver software over) . Apple gets utility for more than just one hyper low selling Mac product. And folks with need for more than 4 USB C slots have a win.
Not putting a x8 PCI-e v3 (or better) slot would inhibit the card vendor from making the card. The mac Ecosystem would get less cards. That would be a next loss all around.
Apple's opinion was otherwise. And their opinion carry an overweight sway since they make final call on design decisions.
Dogma commitment to standard components means Apple couldn't use their own SoC. That is pretty much doomed over the long term.
Some of the non Processor/SoC components flow out of the design decisions that Intel made. Apple is not locked into the exact same set of constraints. If wanted to make exactly what Intel/AMD were/are making then they'd probably would still be buying their stuff. They are not. Apple is not trying to make everything for everybody. Keeping everybody as happy as possible is not their number one objective. There is only a subset of folks they are targeting. It is a big subset, but not 'everybody'.
'real pros' and folks whose business plans are wholly contingent on buying up used e-waste from datacenters to get 'real' work done is probably not a large intersection.
The Xeon SP gen 2 processors don't fully work with the W 3200 series socket. There is a difference in PCI-e provisioning.
"... Max CPU lanes 48. versus. 64 ... "
Intel product specifications
Intel® product specifications, features and compatibility quick reference guide and code name decoder. Compare products including processors, desktop boards, server products and networking products.
Relatively few data centers using the W3200 series to run massive amounts of workload.
As long as Apple keeps putting out reasonable macOS on Intel updates over next 4-5 years , there probably isn't going to be a huge stampede off the Mac Pro 2019 deployments. There will be some vocal , trendy stuff from newest shiny to newest shiny. But lots of 'real pros' have depreciation schedules to follow and the systems aren't going stop being able to do work. ( I'd suspect the flow into retirement of. 2009-2012 back-ups and secondardy machines to be at least as high as the MP 2019 flowing out over the next 1-2 years. MP 2019 slides into secondary machine as the AS systems incrementally roll into the userbase. )
Apple 'sat' on the MP 2013 for nine years. There was a time where a Xeon W3300 (Ice Lake) made sense but that window is closing. AMD releasing the Threadripper 5000 ( e.g. 5995WX) on same day as the Studio launch is indicative that Intel/AMD are just heading is substantially different directions than Apple is.
The Xeon W3300 was marginally competitive in 2021 but as get to the end of 2022 it is just not going to be.
( Dell , HP , and Lenovo has passed on it also. Not just Apple ). At this point Apple's. 10-20 core SoCs are generally beating (or minimally competitive with on some corner cases ) the AMD/Intel. 10-20 core stuff. So not like they are opting for lower performance.
Even if Apple had a 60-90% finished a W3300 stop-gap replacement , the 'gap' is basically gone at this point. ( About 2 months from 2 years from the "about 2 year" transition announcement and about 6 months from Fall 2022 (so even if kick the can of transition to the 4Q 2022 ). ) Unless there is a huge problem with the next generation "double Ultra" Soc for the Mac Pro that would cause a 12 (or more) month delay, the gap is probably too small at this point.
An "intel mac pro refresh" could be. W6950X W6850X Duo speed bumps in the late Fall. If RDNA3 drivers aren't too far of a stretch maybe a W7800X MPX modules in 2023 as a 'last gasp' . Essentially Apple could toss a GPU MPX 'bone' at the market to placate the hyper modular folks. ( decent chunk will buy the mainstream card version that works also. )
That there is an pragmatically integrated GPU ... That is probably inescapable.
Apple doesn't really do standard generic LPDDR4/5. What Apple does is much closer to a semi-custom 'poor man's ' HBM2-3 memory than generic LPDDR. It is much wider and to a non standard RAM packages. That brings it into similarity with HBM technics only the bus and RAM packages are more affordable. But soldering them done is pragmatically also required.
Apple is getting very competitive aggregate bandwidth out of their Memory subsystem. It isn't like generic DDR4/DDR5 is going to be faster. Could burn more power to get to larger Max capacity thresholds but not loosing anything throughput.
At this point it isn't like other folks are going down a similar path.
https://www.servethehome.com/nvidia-grace-2xhopper-supercomputer-building-block-at-gtc-2022/
Soldered down LPDDR5X there also and not using HBM3 ( like the "Hooper" GPU in other package combos).
Is Apple going to use the same kind of outer area and mount points for their package? Probably not. (the variety of systems will be constructing at the high end will be probably be just one. )
All of this is noise. If the new Mac Pro has no slots, I'm out. A Mac Pro with no slots is a failed trashcan mac waiting to happen again.
Apple doesn't really do standard generic LPDDR4/5. What Apple does is much closer to a semi-custom 'poor man's ' HBM2-3 memory than generic LPDDR. It is much wider and to a non standard RAM packages. That brings it into similarity with HBM technics only the bus and RAM packages are more affordable. But soldering them done is pragmatically also required.
Apple is getting very competitive aggregate bandwidth out of their Memory subsystem. It isn't like generic DDR4/DDR5 is going to be faster. Could burn more power to get to larger Max capacity thresholds but not loosing anything throughput.
At this point it isn't like other folks are going down a similar path.
![NVIDIA-GTC-2022-Grace-CPU-Superchip.jpg]()
https://www.servethehome.com/nvidia-grace-2xhopper-supercomputer-building-block-at-gtc-2022/
Soldered down LPDDR5X there also and not using HBM3 ( like the "Hooper" GPU in other package combos).
Is Apple going to use the same kind of outer area and mount points for their package? Probably not. (the variety of systems will be constructing at the high end will be probably be just one. )
I still think Apple will use LPDDR5X SDRAM for the ASi Mac Pro, and it looks like it could also be ECC...?
Probably not.
If the Mac Pro is based on M2 then there are two paths there.
M2 is coupled to A15 ( TSMC N5P ... i.e., something not bleeding edge ).
M2 is coupled to a new TSMC N4 ... and really not all that much significantly faster. ( better power utilization and/or bursty relatively limited clock uplift ).
For either one of those probably not. Remember in aggregate Apple needs hundreds of more TBs of LPDDR RAM as bill of materials than Nvidia would need if they roll out LPDDR5X across the whole M2 line up. The Nvidia Grace isn't even coming until early 2023. It can also pick a more stratospheric memory component bill-of-material cost also. ( Nvidia isn't really selling these on the "street"... they are being stuffed into 'supercomputers'. )
If the Mac Pro is based on a M3 and really a circa 2023 system, then there is a better shot that Apple has moved a LPDDR5X memory subsystem to roll out across the whole M3 line up. [ I'm relatively skeptical that Apple is going to 'fork' a memory subsystem just for the Mac Pro and nothing else in the Apple line up. At least for the base primary RAM. ]
As far as ECC. Either Nvidia is playing 'fast and loose' with the bandwidth numbers or they have cooked up something pretty proprietary (or it isn't really full ECC). LPDDR5 doesn't do full ECC in a bandwidth efficient preserving way.
The M1 series moved from LPDDR4 to LPDDR5 quickly , but that likely was because Apple's semi-custom Memory packages were treating the LPDDR4 die pretty much like there were LPDDR5 dies ( just at a slower speed). Apple's 'uptake' on using LPDDR5 wasn't bleeding edge quick. ( Apple needs high and predictable delivery volume and fixed costs on RAM. Which is way they are generally not absolute bleeding edge adaptors on their silicon designs. )
Finally, Apple's semi-custom RAM packages run more concurrently active RAM dies inside the RAM package than standard LPDDR5 modules do. More concurrent active dies will run into a thermal/heat issues quicker than just accessing a wider set of non-concurrent banks. Running the memory clocks faster if not coupled to a power consumption reduction isn't really going to help Apple's set up. Pretty good chance many 1st generation LPDDR5X implementations just run hotter; not more efficiently. ( for which Nvidia may compensate for with just physically bigger RAM packages ).
Seems like I misunderstood what MPX meant, I didn't realize it was specifically the 2 connector PCIE cards. Regardless, in my head I was thinking of Apple's PCIE cards in their custom enclosure with the big heat sink and mounting screws designed specifically to bolt into the MP chassis. My guess was if the new MP has slots, it will be with similarly designed but smaller cards using a single proprietary connector. I'm not too familiar with the 2019 cheesegrater MP in terms of third party cards, and wasn't aware that there were effectively no third party offerings. Based on what we've seen in the Mac Studio, I'd wager if the new MP has any slots/upgradeability it will be some proprietary Apple solution, which is why I suspect it won't have any standard PCIE connectors. I also doubt these slots will support discrete/additional GPUs since Apple seems dead set on the integrated die approach of combining CPU/GPU/RAM into one SOC, and almost certainly won't support slotted discrete GPUs. I don't know enough about integrated SOCs to say if it's technically possible or not, I'm just guessing they won't do it (unless they can charge absurd $$$ for it, assuming it is possible).
I worded this poorly. I meant that in ~5 years, most of the "real pros" using the 2019 MP for real work will probably have moved on to a more capable/modern system - whether the replacement is AS or Windows/Linux remains to be seen. When these work/business machines are liquidated, they will be bought up a variety of people, ranging from "prosumers" to "real pros with less budget" that still want to use them ("prosumers" being the type of people who aren't using it for a job or "real work", but still want "pro" or unique/different hardware, even if it's a little dated). This happened with the previous MPs. Even the 2010 MP is bought/used by people, judging by the amount of MP 1,1-5,1 posts on this forum. I obviously can't quote numbers, but I have seen pictures/videos on YouTube and other places of 2013 trashcan MPs in use today ranging from small music studios, to small photo/video studios, to people like me and others who just like the design and don't mind using older hardware since we aren't using it for business/work purposes.
Anyway, in ~5 yrs the Xeons that are compatible will have likely have fallen in price enough to be a relatively cheap upgrade. Though from my understanding this depends on whether or not said Xeons were used in datacenters or similar places and are liquidated en masse like the 2013 MP Xeons. I don't know the technical terms but for example, the Xeon E5-2697 v2 (12 core highest tier Xeon for 2013 MP) can be had for $80 to $100 on eBay now. I don't know the original price from Apple or from Intel but it's certainly at least 10x that. Already, the Xeon W-3275M (28 core used in the 2019 MP) can be had for ~$1200 on eBay, and there are quite a few low-spec 2019 MPs selling for $3000 or less in the last few months. Extrapolate this out a few years, and the 2019 MP could be relatively affordable if someone really wanted it, though I'll be the first to admit using it for a business/work application probably isn't the best decision.
Sorry for any confusion, a lot of this isn't in my area of familiarity.
...
Dont be sorry. You're being fed bullying obfuscating bs. There are TONS of 3rd party options/cards for the 2019 MP. I'm using PC (non-apple) Sonnet 6900XT card. It works better than the Apple Card and is way cheaper. I could have bought probably any of a dozen versions of that card from a bunch of different 3rd parties. Then there are the 6800 variants and many other variants. I pulled the PoS apple 580x from the machine and also use an AMD9100 card. You can add ton of 3rd party audio cards, USB cards. Most of these are made for PC with no need for drivers, and they will just work on the Mac. Im using a PC Highpoint 7120 card that lets me attach and use U.2 format drives (currently using a 15.36TB SSD). There are no drivers, and this is a PC (not Mac) card, by a 3rd party and works great. There are also tonnes of 10GB copper and fiber channel network cards that do 100GB-400GB throughput! PC only cards that just plug and play in a slot with no drivers. I have one. Works great.
The MPX cards, while nice and are Apple only, really are basically irrelevant. They let you supply additional volt/amps without needing a cable so if youre OCD like apple, that's nice, but hardly necessary. They also provide a little extra bandwidth so you can route/map some extra thunderbolt ports. But again, while that's nice, completely unessary. The MPX slots are really just a regular PCIe slot, and another MPX slot behind it. You can just ignore the MPX part and just shove in some non-apple even PC/only card and ignore the MPX part of the slot and it will work just great.