With all due respect...too much bad blood, too many egos involved on both sides of the negotiating table and too much competition with the desire to shape the future of Machine Learning trumps any post on the darknet.
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 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
With all due respect...too much bad blood, too many egos involved on both sides of the negotiating table and too much competition with the desire to shape the future of Machine Learning trumps any post on the darknet.
Darknet's track record on this thread from the beginning is more of a punchline than credible. But even if those two worked out their differences the "need" for an MPX module with NVidia would be limited. Doubtful Apple would put much effort into doing multiple vendors in "AMD vs Nvidia" line up fashion. Apple doesn't need Nvidia for an entry card. Apple doesn't appear to be doing anything in the "middle" range at all ( with AMD or not).
Unless new Titan RTX is using HBM2 a Duo would be problematical. GDDR6 is cheaper but it also takes up way more logic board space. Apple's Duo card is pragmatically enabled by using HBM2 and space more efficiently. It costs more but it meshes with direction Apple is going with other embedded GPUs across the Mac line up. Apple is in position to push the volume here until costs get better longer term. ( won't turn into the cheapest option but more viable over larger set of systems ).
I think it's a political move by Apple. They know that people won't trust that Apple will ship more MPX modules, so they want to make a show of it taking generic PC graphics cards to reassure people that their purchases will stay relevant no matter what.
There are edge cases of specialty GPUs that Apple may not supply MPX modules for too. Machine learning specific GPUs, etc. Nvidia could even fall into that.
They both use the Thunderbolt protocol. The one major substantive difference is that DP v2 changes all the channels to be outbound channels. Thunderbolt is bidirectional traffic and DP v2 is unidirection traffic. But the format the data is being sent in is the same. It is just the direction of traffic on the "lanes".
So yeah 4 lane concrete highway is a lot like a two lane concrete highway. Some places in a evacuations for a natural disaster the direction on the highway is changed to go 4+ lanes out and zero in. The highway didn't get "different' with the directional change. Same kind of relationship between TBv3 and DPv2.
Thunderbolt (TB) needed a connector to piggy back off of so linked up with VESA to help get more traction for DP while at the same time growing TB. TB made another deal with USB-IF to get Type C and alt modes. ( and faustain bargain there was that USB 2 had to coming along for the ride everywhere. Funny how standards body's want their standard to move forward just as much if doing a deal with them. ). That enabled and pulled TB and DP into alt mode (and get Type C added as a standard port for DP. )
No. DP 2 is 80 Gb/s no 40. And the '40' of USB 4 is essentially TBv3. So yeah that would make them highly related too. It isn't 100% clear yet what 'mutation' USB 4 will put on TBv3 but again is the same protocol being redeployed.
( USB 4 might make the DP streams more optional. That wlll probably further muddy the waters of end user confusion over what Type C port does. "Cable fits, How come it doesn't do X? ". But that will make the folks on USB-IF who are on race to the bottom implementations happier so it will probably come. )
If Thunderbolt had created its own new connector it probably would have failed. Fact is that a portion of Thunderbolt agenda was Display Port's agenda. That is the deal they worked out to get to "hijack' the port that was already out there.
Again not really pragmatically true. Intel and Apple contributed well over 50% of the folks who worked on Type C extension to the standard. If Intel and Apple hadn't put resources in to Type C it probably never would have happened. USB-IF probably would have capped out at 10Gb/s as the "it has to be cheap"/"it has to have limited physical change" folks would have won out on the voting.
No, I largely mean the protocol. There are some interactions with speeds in that there are some presumptions in the protocol that will employ active connections for higher speeds to achieve longer lengths. And that there should be an straightforward path to putting in somewhat affordable fiber.
DPv2 is also picking up the restrictions that TBv3 on active cabling for distance that DP and USB-iF have largerl eschewed. That basically is linked to the transfer speed. You just can keep cranking affordable copper up in clock/frequency and not 'pay' a price on distance (and/or errors/interference creeping in).
Some people complain about change no matter what it is. And some folks will complain because it knocks out their own vendor lockin ( e.g., PC system vendors each with their own docking station connector. ).
Taking a working system and adding some more significant new set of features to it is very often less risky than creating a brand new complex system from scratch. Where Thunderbolt is no is pretty close to where they wanted to go in the first place. ( Remember the initial stab in lightpeak/thunderbolt was trying to leverage the USB socket and was rebuffed by USB-IF. Sony had earlier evolution devise with USB sockets ). Thunderbolt built momentum by linking up the DisplayPort. Later took another stab at merging in with USB but into a new, for future enabled port and it work. So that momentum made a substantive difference.
Thunderbolt is mature enough at this point that letting in more folks have access to it will more likely lead to better incremental diversions. Early on in a standards development too many "cooks in the kitchen" can screw a standard up. Or make it hard to come to a consensus if there is too many conflicting views. ( e.g., OpenGL was 'largely' stuck until AMD donated Mantle to be starting point for what became Vulkan (glnext) . )
But there is nothing inherent in macOS that absolutely requires that applications can only be in the application folder of the boot drive. Some apps may have their own screwed up restricdtions due to copy protection hocus pocus or wonkly hardcoded paths, but that should be tagged on them as to why they are forcing their users to buy hyper expensive Apple SSDs.
Sure their can be "big pig" apps but they can be put on their own drives by themselves. Putting a huge pile of static data that isn't even yours into your normal backup stream is a bad idea. It tends to bloat out the whole back-up system. Two static copies of huge static data is enough.
And he would STILL be wrong because there are separate power plugs aside from the MPX slots for power. All you have to do is literally look at the motherboard to see them.
[doublepost=1562176070][/doublepost]
We do not know about the boot screens, but yes, regular PCI cards will work. Beyond SSD cards etc that have no boot screens, even video cards work on the 2010 cMP. In fact, that is how the 2010 machine is still compatible with Mojave, by using 3rd party screens, sadly, many without boot screens. So saying they wont work because of proprietary MPX cards is just completely wrong.
It's a good and fair question if Apple will make the boot screens work better on the nCMP. Time will tell.
[doublepost=1562176524][/doublepost]
I plan on using the WX 9100. It drives 6 minidisplay port screens off one card. Will be good to drive my now ancient 30" cinema displays. Maybe someday apple will make the monitor everyone wants. A $1000 5k 27" display off the iMac "for the rest of us".
Worth noting the onboard MPX module can drive six DVI displays as well. Unless you're really saving a buck by using a WX 9100 (won't know until the BTO prices drop), the Vega 2 would look to be a better option that can drive the same number and same type of displays.
[doublepost=1562180344][/doublepost]
I have sources within Apple (not some dark net guy) that have told me off the shelf PC Cards will be supported, as long as drivers are present. It's a configuration they specifically considered and built for. You will lose video over the built in Thunderbolt ports unless you keep an MPX card in one of the other bays.
I don't know anything specific about boot screens.
The issue isn't whether one can get the same specs for less money. The issue is whether one can buy an internally expandable, headless system for less. The answer is: Yes, you can.
[doublepost=1562197195][/doublepost]
Agreed. But that's not what you faulted him for.
@deconstruct60 is suggesting the MPX modules (presumably the cooling shroud) might physically block cable access to the power connectors, even if the system doesn't power them down when an MPX module is detected.
The MPX power slot probably shares the same power source as the respective 8 pin connectors. MPX probably covers up the 8 pin so you can’t use both at the same time.
that would be a simpler solution, than building in a switching system to physically disconnct power from the sockets when an MPX is installed.
In a few years, dremelling a small chunk out of the MPX cooling shroud will be the new pixlas mod
It isn't really a "suggestion". All folks have to do is actually look at the pictures on the Mac Pro overview , design, tech specs page.
MPX Bays empty ( dual 8-pinpower sockets horizontally mounted (horiz relative to logic board not these following pictures ) in the middle (just a bit below the second double wide socket line of the of the top/2nd MPX bay. Just slighlty above on the first MPX Bay. It is nudged slightly better on the 1st Bay, but are in the general middle of the bay. The rectangles on left hand edge with the " ! in a triangle" icon on them. )
MPX Bays full with Full sized MPX modules
Can you see the horizontal dual 8-pin power sockets? No. That isn't a suggestion. You can't. So the access is obstructed. If you pull the uppermost module out of Bay 2 then the that bays power would be unobstructed.
Additionally, note that the retention bar for the MPX modules on the left hand side also goes directly over the space where the power connectors are.
I not proposing that there is something about the shroud that reaches out and drops right on top of the socket and snaps into place. The visual evidence right here though factually shows that getting to the power sockets is very obstructed. It is an obvious, common sense clue that since covered it up and made it very hard to get at so it is in a "don't use this" status.
The fact that Apple placed the power connection in the middle of the bay means they were trying at all to provide access when full size modules are in place. It is pretty hard how to come put another place that could hide it any deeper out of the way than that.
Even for a "half" height 580X module you still won't be able to see most of the socket. Maybe the top of the horizontal connector is close enough to the top edge of the shroud to get to but if try to stuff a double wide card into the x8/x16 upper slot of the bay how is that card not now in the way?
The way that will work that probably won't cause problems is for a x16 double wide card in the "lower" x16 slot in the MPX back and ether nothing or a modest (or 'half lenght) card in the 'upper' slot of the MPX bay. At that point the double 8-pin power socket is reasonably accessible. If try to pack in very long , double width cards into the four slots (over the two Bays) then that would have similar problems to the full size MPX modules.
Any notion of trying to 'snake" power cables under the full sized MPX modules to the power slot in the gap between the shroud and the logic board is highly dubious. The shroud is going to get hot. So hot metal on your plastic cord is going do what? Probably not a good outcome. Pinning the power cord to the logic board. Again Probably not a good long term outcome.
Even with the half height MPX module the hot shroud will be in close proximity to the cables coming out of that socket. I suppose some folks could 'invent' something that was suppose to get around that, but common sense is that if cover from direct view probably means "don't use this". Both half and full sizes modules cover this from view because of its middle placement. That is probably not an accident. Some "working space" is needed to use those.
There are no pictures in place or explicit dimensions for the Pegasus R4i
https://www.promise.com/us/Promotion/PegasusStorage
but it looks like a full size MPX module. ( the tongue with the two screw holes on the left match up with the top MPX Bays 2 module above. ) . The enclosing sheet metal of that probably won't be problematically hot, but it is still going to obstruct access to the power ports. Four full sized 3.5" that need to be somewhat vibrationally isolated is going to take some space. Plus going to want to let enough air get though so super dense vertical packing isn't a good idea.
If this card doesn't plug into the MPX Connector slot there may be a way to snake out out power cable(s) to some other card, but it won't be straightforward. ( if juggle things "just right" on start-up spikes then 75W to run 4 HDDs and a RAID card logic board could work. ). But that will probably be a "Frankenstein" corner case;
P.S. If plug in Apple's Afterburner card into slot 5 the 6-pin power port isn't all that great to get at either. That one though is offset enough so that the inputs to the power sockets are a bit above the slot space that relatively thin card will occupy. The Afterburner card probably won't be happy with really hot neighbors on both sides that are relatively hot.
Last edited:
It isn't really a "suggestion". All folks have to do is actually look at the pictures on the Mac Pro overview , design, tech specs page.
MPX Bays empty ( dual 8-pinpower sockets horizontally mounted (horiz relative to logic board not these following pictures ) in the middle (just a bit below the second double wide socket line of the of the top/2nd MPX bay. Just slighlty above on the first MPX Bay. It is nudged slightly better on the 1st Bay, but are in the general middle of the bay. The rectangles on left hand edge with the " ! in a triangle" icon on them. )
View attachment 846636
MPX Bays full with Full sized MPX modules
View attachment 846637
Can you see the horizontal dual 8-pin power sockets? No. That isn't a suggestion. You can't. So the access is obstructed. If you pull the uppermost module out of Bay 2 then the that bays power would be unobstructed.
Additionally, note that the retention bar for the MPX modules on the left hand side also goes directly over the space where the power connectors are.
I not proposing that there is something about the shroud that reaches out and drops right on top of the socket and snaps into place. The visual evidence right here though factually shows that getting to the power sockets is very obstructed. It is an obvious, common sense clue that since covered it up and made it very hard to get at so it is in a "don't use this" status.
The fact that Apple placed the power connection in the middle of the bay means they were trying at all to provide access when full size modules are in place. It is pretty hard how to come put another place that could hide it any deeper out of the way than that.
Even for a "half" height 580X module you still won't be able to see most of the socket. Maybe the top of the horizontal connector is close enough to the top edge of the shroud to get to but if try to stuff a double wide card into the x8/x16 upper slot of the bay how is that card not now in the way?
The way that will work that probably won't cause problems is for a x16 double wide card in the "lower" x16 slot in the MPX back and ether nothing or a modest (or 'half lenght) card in the 'upper' slot of the MPX bay. At that point the double 8-pin power socket is reasonably accessible. If try to pack in very long , double width cards into the four slots (over the two Bays) then that would have similar problems to the full size MPX modules.
Any notion of trying to 'snake" power cables under the full sized MPX modules to the power slot in the gap between the shroud and the logic board is highly dubious. The shroud is going to get hot. So hot metal on your plastic cord is going do what? Probably not a good outcome. Pinning the power cord to the logic board. Again Probably not a good long term outcome.
Even with the half height MPX module the hot shroud will be in close proximity to the cables coming out of that socket. I suppose some folks could 'invent' something that was suppose to get around that, but common sense is that if cover from direct view probably means "don't use this". Both half and full sizes modules cover this from view because of its middle placement. That is probably not an accident. Some "working space" is needed to use those.
There are no pictures in place or explicit dimensions for the Pegasus R4i
https://www.promise.com/us/Promotion/PegasusStorage
but it looks like a full size MPX module. ( the tongue with the two screw holes on the left match up with the top MPX Bays 2 module above. ) . The enclosing sheet metal of that probably won't be problematically hot, but it is still going to obstruct access to the power ports. Four full sized 3.5" that need to be somewhat vibrationally isolated is going to take some space. Plus going to want to let enough air get though so super dense vertical packing isn't a good idea.
If this card doesn't plug into the MPX Connector slot there may be a way to snake out out power cable(s) to some other card, but it won't be straightforward. ( if juggle things "just right" on start-up spikes then 75W to run 4 HDDs and a RAID card logic board could work. ). But that will probably be a "Frankenstein" corner case;
P.S. If plug in Apple's Afterburner card into slot 5 the 6-pin power port isn't all that great to get at either. That one though is offset enough so that the inputs to the power sockets are a bit above the slot space that relatively thin card will occupy. The Afterburner card probably won't be happy with really hot neighbors on both sides that are relatively hot.
Can you see that the 8 pin connectors are along the plane of the the motherboard, in the photo pointing up. You put a ribbon in that, it goes up flat under the slots. They are available. This is noise.
[doublepost=1562261166][/doublepost]
I criticized him for several things. This among them.
I already have a 9100 laying around, so it works out for that. Do you know if all the MPX modules have 6 DVI ports? Thanks so much.
don't necesarily need Vega II MPX modules. The 580X module will drive them also. Have to consume the some (or all) of the standard Thunderbolt ports, but it would drive them. If using direct adapters only 5 ports come directly off the MPX Vega II module. Still would need to grab one from one of the Thunderbolt ports ( And the solo Vega II only drives just one of those Thunderbolt pairs. Top or I/O card ? )
If one already has 6 mini-DisplayPort to dual link DVI adapters already it isn't really just the cost of the respective GPU output solutions though. It is also 6 new adapters. Even at $25 a pop thats another $150. Plus it is two different types of adapters (if use the HDMI connector on Vega II ).
The WX9100 has some substantive discounts now. Newegg has it at $1,499. Apple/AMD beating ~$1349 is a bit doubtful. Apple/AMD beating the WX91000 initial price of ~ $2,100 has a decent chance. Also with WX9100 along side the entry MPX module would have 6 more screens. Not basicall swapping 6 port for 6 ports and have two open x8 slots for storage / I/O / etc. (Vega II fills up a whole MPX bay ). The opportunity cost for those open slots if probably worth another couple $100 to more than few folks. ( A major selling point for the base price of the system being $6K is basd on slots to use. Full MPX module takes two. That might work if using the Thunderbolt ports are thunderbolt ports. But if if purely in backward capability DP alt mode .... that is a 2nd wasted slot. )
That was substantively also true of the 2009-2012 era Mac Pro's also and that didn't lead to any random card being able to work either. If Apple goes into full Rip van Winkle mode on their portion of getting new drivers done and shipped then it isn't particularly any different outcome than if they go Rip van Winkle on MPX modules also. The notion that Apple can be 100% disinterested in 3rd party GPU subsystems and progress not be impeded in any way is a bit fanciful .
apple card for boot screen and 2nd augment card for "bigger" GPU was a common mode of the Mac Pro slot survey done here on this subform. I won't put it past Apple to "hand wave" supported by driver here as in post boot phase and happens to work once full OS.
There is a certain faction though that only want to use non Apple GPUs; evicting everything Apple possible from the Mac Pro. I suspect that was not an support configuration they were looking to fully enable. The fact that many folks who specifically ask the question makes it something that Apple can likely just tap dance around.
The direct edge connectors on the Vega II MPX GPU modules so far are
" ... Four Thunderbolt 3 ports and one HDMI 2.0 port on card ..."
https://www.apple.com/mac-pro/specs/
For the entry card even fewer.
" ... Two HDMI 2.0 ports on card ..."
All the display formatted output on that card goes out to feed the Thunderbolt controllers. I think the edge of that card is totally aimed at moderately priced 3rd party monitors. It isn't aimed at legacy Apple monitors at all. [ Which is probably a good thing since most of the historical Mac Pro base probably didn't have Apple monitors. ]
Since most of the "edge" of the MPX modules is for moving Air out of the system, I suspect 5 is probably the cap for now. Maybe when DP v2 comes along , a newer GPU with more outputs and Apple swaps out the HDMI they could sub two Type-C connectors for the one HMDI.
All three photos are oriented in the same direction.
Trying to 'snake' a ribbon cable under the full size MPX modules is unlikely to be flat. First, you have to traverse a distance of two standard slot width to get from under. the full size module. Second, where the ribbon connection is highly likely to go to after that is up vertical tangent the motherboard. Both of those pragmatically combine into a likely situation where the cable is bowed, not flat, and that the cable comes into contact with one or both ( logic board and/or MPX shroud ). The only 'noise' here is that is a good idea for a practical context to put that cable into.
The half width MPX module really isn't that much better. It needs to go vertical even more quickly if trying to snake up between the cards or needs to pass under and then up the card in the 2nd MPX bay slot.
Whether you should stick something into a socket is more important than you can possible stick something into it. "It will work because it is physically possible to jam something in there" is not a well grounded interpretation of what is visible in those pictures.
Hopefully Apple distributes some clear support documents ( and maybe even write a user manual ... what a concept? *cough* ) before they ship these system. If Apple is solely relying on common sense then they may have some problems. [ Although, I'm sure that won't mind applying a 'Darwin' tax on those who want to fry their systems on their own dime and need replacement parts. The motherboard and power supply are going to cost way more that the "I could build this with my trusty screwdriver and thermal paste" bill of material lists tossed around on these forums. ]
Conceptually someone could build a low heat radiating ( non MPX connector power) MPX module with a ribbon cable guides to control the bowing. However, that wouldn't be in conflict with Apple's likely "only use one of these options" constraints. What Apple has built extremely likely is not in that category though. Did the Engadget commentary account for that. No? Has any of the MPX module builders so far accounted for that yet. I'd be very surprised.
The 2012 Mac Pro wasn't supported for standard PC video cards beyond having PCIe slots. It wasn't a configuration officially supported by Apple. Some important functionality, like firmware updates, wouldn't even work with a PC card.
All the Thunderbolt ports can double as DVI ports. Including the ones on the machine itself. USB-C to DVI adapters are plentiful and look pretty cheap. A one Thunderbolt to two DVI adapter might even be possible.
In other words it was the drivers ( software/firmware) that was the critical piece. Essentially the same thing now. The gap is incrementally close with UEFI GOP but the 'standard' on the PC side often really isn't complete compliance with the standard.
the "pretty cheap" USB-C to DVI adapters are not dual-link (at least dual link at the old 30" resolutions). Apple's not so cheap HMDI -> DVI isn't either.
Ask Aiden if he knows somebody waiting to buy a ncgMP with 4 Titan or Quadro RTX on board, about DNG, you're right about these speculation, but there are few facts: CUDA's repositories have been getting more patches related to macOS the past 6 months than last 2 years.
Disagree, there are N ways to fit 2 Titan RTX between the MPX real state.
With all due respect, neither apple belongs to Frederighi, Cook, etc neither nVidia belongs to Huang, both belongs to their stock holders and theirs market period.
If you believe those stories about a multi-billion corporation being managed as garage.grocery you also believe in Santa.
I think those stories are mockup by media linked or apologetic to some people at Apple, but while going all on AMD GPU had a business case due costs, banning nVidia from the Mac ecosystem has become too expensive and in unsustainable, is something Apple managers can't withhold too long, as nothing in AMD, Intel or ARM is even close to be competition or moreless viable alternative to CUDA, i come from being dedicated to HPC application development before on scalar only CPU I tried to jump bto opencl/AMD and failed because opencl is 2 or 3 years behind Cuda on both language/platform features and performance as with the toolchain, despite efforts like vulkan, vuda,, SYCL, it's like compare basic with c# or python, AMD is barely getting close (on 7nm) to rtx 2070 performance while nVidia (on 14n) next year when next nVidia architecture arrives with 7nm PCIe 4 and what ever, AMD will be again the loser well behind the leader, even Intel do not plan vto fight nVidia NeXT year they are aimed at same niche as AMD: low end.
I'm very confident there will be nVidia drivers soon in Catalina and at least support fot std or oem rtx GPU all these courtesy of CUDA/tensorflow.
Even i believe (this is my personal speculation) the industry could adopt the electrical speciations of the mpx module for non-apple motherboards. (Thunderbolt, power and 16 extra (optional) PCIe lines).
Note as yet Apple hasn't filed for mpx patent neither alike, while at least not publicly they do not reveal it's specifications.
Apple choosing to keep NVIDIA off of their platform has everything to do with AI and machine Learning and zero to do with performance of CUDA versus OpenCL or an RTX 2080 versus a Vega 2 Duo.
Apple is staying platform agnostic and looking out for its own best interest and allowing NVIDIA GPUs on the Mac and macOS platform is not in their best interest.
Apple has a SVP of AI and Machine Learning (John Giannandrea - https://www.apple.com/leadership/john-giannandrea/). There are only 12 other people in the SVP role at Apple, which should tell all of us just how important they consider AI and machine Learning. This a now a core business technology to Apple. There is zero chance they will let NVIDIA in the door to subvert their plans in AI and Machine Learning to give you, me or anyone else a faster GPU.
CUDA is proprietary...it matters zero to Apple that it is the leader in the market and you're deluding yourself if you think Apple cares. Apple builds its own CPUs and GPUs now and if they move the Mac to ARM, then AMD GPUs will be there for a while, but I suspect even they will give way to Apple's own GPUs. Apple letting NVIDIA GPUs on the Mac nets Apple nothing and it doesn't jive with Apple's goals in having developers, specifically iOS developers learn and use CoreML building applications that utilize on device Machine Learning on the iPhone and iPad and then possibly the Mac. In those scenarios, NVIDIA has no viable place in the Mac ecosystem.
Besides, all of NVIDIA's CUDA tools are on Windows, so what would be the point of having NVIDIA on a Mac Pro? The last CUDA driver is limited to High Sierra and even if the toolkit was recently updated, I don't see CUDA on macOS being a high priority for NVIDIA either way. It's just a distraction for both companies.
Apple belongs to the shareholders...but they seem to trust what Cook and Company are doing and I am 100% sure that they are not going to give NVIDIA and Huang any chance whatsoever to derail their AI and Machine Learning plans. They are too intricately tied to the growth and success of the iPhone, iPad and iOS...which at the end of the day, are the only things that matter, not whether a few devs can run an RTX 2080Ti in a Mac Pro.
Were I running Apple, I would do the exact same thing. NVIDIA GPUs run on Windows and that should cover most of the market, but in no way would I let NVIDIA anywhere near my platform...they are not a good partner, they are a market competitor, pure and simple.
I can almost guarantee it...it will also disconnect the adjacent PCIe slot as well, as there aren’t enough lanes to drive it, so I expect that any Apple MPX module will dynamically/automatically turn off Slots 2 and 4 if 1 and 3 contain an MPX module.
I see people already trying to subvert Apple’s designs thinking they can get around what Apple has engineered and they are all going to be very disappointed when they either can’t find a workaround or they blow up their minty fresh Mac Pro within the first couple of weeks. I bet 90% of the delay in getting it shipped is the documentation team needs more time to read about all the scenarios being hatched on this site, so they can write warnings and tech notes for the would be hackers here.
Any one with a reasonable grasp of mathematics should be able to work out that there's sound reasoning behind blocking off those 8-pin connectors when the MPX bay is filled (and if/when OWC and others come up with third-party solutions, they'll do the same). Say, for example, Mr Croesus buys his 28-core system with a couple of Vega II Duo modules installed, and (in an unexpected tightwad moment) ponders if he can chuck the Vega Frontier Edition he was running in his old cheesegrater in the new one. No, because Apple doesn't want his PSU to shut down/his fuse box to trip/act of god to ensue, and has therefore blocked off those 8-pins with the MPX modules. He'll have to make do with some eGPUs…
Let me contend your biased (by media articles, not reality) analysis:
1st Apple doesn't control neither has hopes to control AI development, AI by now belongs to tensorflow (released by Google, comunity controlled, Apple supports tensorflow-lite)
2-tensorflow it's an framework for AI with TF you can train an model and run (inference) it , you can train a model in a CUDA cluster and do inference with this model (run itl with any other platform as it can run tf or tf-lite.
3 tensorflow strictly do not requires GPUs, but GPUs TPUs FPGAs accelerates inference and training by two magnitude order, what's its Cuda advantage? Many, a single RTX Titane GPU is even more powerful than both Radeon pro Vega II Duo in the cgMP at just 280W and at 14nm process, beyond that tensorflow still do not support Metal only tf-lite supports Metal (for inference only), nVidia positioned it's ai solution when they developed their GPU around CUDA instead to repurpose existing hardware (and Radeon), this enable a lot of programming flexibility with CUDA along superior performance (while gaming doesn't get big benefits from this), but it doesn't ends there, Metal GPU offloading features seems a rebranding of opencl, feels short to provide the features array present in CUDA, it's not just inferior to CUDA ir cant replace CUDA with efficiency beyond the few similar features, it's something people like davinci resolve developers among many are aware, aren fee lines of code to Port, some things needs vto be fully rewrite just to work properly (going from Cuda to hip).
Your assumption about Apple crusade to control AI development is bogus, the best evidence on that is Swift for Tensorflow (Swift-tf) it only supports CUDA (read: Swift, SWIFT I didn't name python).
Apple, or Cook is controlled by both shareholders meetings and the reality, and the truth is with Cuda you have better more efficient AI and you can deploy the latest tensorflow features.
Today only GPU accelerated AI in macOS rely on :
nVidia(macOS<Mojave): tensorflow, Swift-tf, tf-lite or
metal:tf-lite keras (thru plaidML to be discontinued by Intel)
that's all the rest do not deserve to name.
Wthis panorama after 2 years trying to impose Metal over CUDA failing so miserably, what Apple can do? Even their money couldn't buy a full support for Metal in tensorflow (impossible since lacks CUDA flexibility), even having money to develop custom you (using ARMs IP) they have to choose lost 2-3 years to still being behind CUDA or simple allow Mac users to use Cuda at least meanwhile someone develop a macOS compatible TPU (like Google's edge(Coral) TPU).
Did you know even Siri rely on. Cuda/Linux servers?
Sorry with the deserved respect you're wrong, you remember me some guy arguing about Apple wireless charging was delayed because apple will implement long range wireless charging tech, he only readed Apple-apologetic blogs ignoring what physics laws said about wireless power, apple got later to the wireless party because their executive where distracted not because they planned something superior, same here about AI 3 years ago some idiot asked to retire from vulkan (even vulkan it not in better position about CUDA) but even to close nVidia Doors, this cost apple the AI race, now AI developers think on a Linux workstation (even a Windows machine is better suited) for Tensorflow development.
Last edited:
the 8-pins plug into the top edges of those connectors, so they're going to be running flat against the motherboard coming out of the connector. Unless the cooling shroud for the MPX module is flush against the motherboard, sealing the plug in, there's going to be a gap to snake a cable through between the motherboard, and the shroud. While it may become warm, if it's hot enough to damage / effect a cable passing between it and the motherboard, it's probably going to be hot enough to damage the motherboard.
Personally I think it's likely that a sensor in the MPX slot will cut power to the 8-pin connectors when an MPX card is inserted.
Last edited: