With the T2 chip repair protocols in place with the MacBook Pro and iMac Pro. I believe there is 0% chance there will be any user upgradability and repairs to the new modular Mac Pro. Hot swappable bays via thunderbolt and other interfaces are available and Apple figures that’s good enough. 2019 will bring us the cube 3.0. Probably completely sealed in a chamfered glass enclosure with a series of thunderbolt ports and maybe a power input.
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
I don't see that T2 changes a huge amount. Whether the PCIe is internal (a slot) or external (a Thunderbolt connector) isn't really material to T2. T2 isn't linked to the GPU.
T2's disk controller is a possibly sticking point with changing internal drives though. We'll see where Apple lands on that. I honestly think enforcing encryption on the built in drives on a desktop is a little silly.
Absolutely. Internal drives I think will be locked in external not (already third party and apple blessed raid drives etc.). I could see Apple pointing towards those (external options) with a single or dual SSD option internally (locked down).
It's possible a T2 Mac Pro could encrypt primary storage, but have other NVMe slots which are not controlled by T2 and are swappable. But again, the idea of a self encrypting Mac Pro seems like a silly idea in the first place. So I don't know why Apple should even bother making it that complex. They could even just turn off encryption on T2.
It’s possible. But looking at what they have pushed out recently, especially the small details, like the power cord on the HomePod, I think more and more we will be shown a small flush glossy box that can’t be opened by the end user. I’d like to be wrong (seriously I would), but I’m tempering my expectations with what they’ve been doing.
It’s possible that there would be a ‘module’ for hotswappables but even then I think apple has ‘promise’ covering those needs (or at least what apple considers to be covered via thunderbolt).
It’s possible that there would be a ‘module’ for hotswappables but even then I think apple has ‘promise’ covering those needs (or at least what apple considers to be covered via thunderbolt).
I wonder if those people who got invited to that Apple meeting where they assured their commitment to the Mac Pro; if Apple have continued feedback talks to those people? Have you guys heard anything? Or was that meeting just a general meet and greet and then goodbye and Apple going back to locked doors?
I really hope Apple have kept in touch with non-Apple industry people when it comes to the Mac Pro and those people really press Apple about a truly modular Mac Pro.
I would be fine with a Mac Pro built with a "cartridge" system like the Surface Hub 2 will supposedly have. So same enclosure for years and years to come but Apple keeps updating the "cartridge" or "sleds" with new hardware as soon as they are made available from AMD, Intel, Nvidia,
I really hope Apple have kept in touch with non-Apple industry people when it comes to the Mac Pro and those people really press Apple about a truly modular Mac Pro.
I would be fine with a Mac Pro built with a "cartridge" system like the Surface Hub 2 will supposedly have. So same enclosure for years and years to come but Apple keeps updating the "cartridge" or "sleds" with new hardware as soon as they are made available from AMD, Intel, Nvidia,
Yep. I’m not sure it makes sense to expect a user-upgradeable Mac Pro chassis when all the available evidence tells us that that would be fundamentally contrary to Apple’s established behaviour and apparent design and business principles.
[doublepost=1538843566][/doublepost]
My understanding is that meeting was with Apple-favoured members of the press (who are not the Mac Pro target audience), and was a marketing activity, not a forum for gathering feedback on the product.
Been wrote here many times, the mMP will be "modular" as by Apple understand: Propietary, Apple-sourced or Apple-certified Upgrades, minimal DIY upgrades (as much RAM and Modular-NON-STD GPUs) no Internal Storage tobe DIY upgradeable (only by Apple or authorized Shops).
I'm moving Away, I'm on linux now, AMD TR + nVidia Titanium GPUs.
All my favorite tools run on linux: PyCharm, CLion, Golang, CUDA 9.
Even I'm considering a mini-cluster with 4 AMD TR+8 RTX GPUs+4 100G Lan, should be less expensive than the foresee maxed out mMP and at least 4x more powerful.
Only miss few macOS goodies, as Desktop integration,, no big deal, Linux still has Mani quirks, and is no choice for Media/CAD Workflows, but for Compute, HPC R&D has no rival.
I'm moving Away, I'm on linux now, AMD TR + nVidia Titanium GPUs.
All my favorite tools run on linux: PyCharm, CLion, Golang, CUDA 9.
Even I'm considering a mini-cluster with 4 AMD TR+8 RTX GPUs+4 100G Lan, should be less expensive than the foresee maxed out mMP and at least 4x more powerful.
Only miss few macOS goodies, as Desktop integration,, no big deal, Linux still has Mani quirks, and is no choice for Media/CAD Workflows, but for Compute, HPC R&D has no rival.
This is it I think. The T2 is now acting as a killswitch against any hardware intrusion on the mbp, no reason not to expect it to be such on the next macpro. Even if it does come with an internal pcie slot, unless you are rocking some particular Apple software tool, opening the box to insert a GPU will more than likely temporarily brick the box until a genius can run the software to unlock it again.
Last edited:
I think Apple will direct you to an epgu, no need to open up the ‘cube 3.0’ just add another box from another company.
Apple's comments on GPU throughput make me think otherwise. While eGPU/Thunderbolt throughput is good enough for most, that wouldn't match up with Apple's comments on wanting to support the most demanding GPU workflows. So I don't really see eGPU/Thunderbolt being where Apple goes as a GPU bus link.
I'd give decent odds that them playing with all these third party GPUs for Mojave on the Mac Pro 5,1 is them possibly piloting off the shelf GPU support for the next Mac Pro.
Might be why Apple's bug response have shown interest in cards like Vega that shouldn't even be recommended on a Mac Pro 5,1 for power reasons. Mac Pro 5,1 users could be beta testing changes for the next Mac Pro already.
(Not beta testing actual firmware obviously. More like testing drivers against PCIe cards.)
but locking the mac pro to 1 ssd under the pci-e X4 will suck even more so if they put 2-3 on it.
https://www.techpowerup.com/248305/...terface-for-gpu-compute-on-vulkan-open-source
WHO WOULD HAVE THOUGHT?!!!111one1oneone1
WHO WOULD HAVE THOUGHT?!!!111one1oneone1
This is primarily just hysteria more than fact. The "intrusion detection" isn't the core issue. Security subsystem tamper resistant is far more like the driver here. ( e.g., Servers being tampered with before deployment stories. https://www.macrumors.com/2018/10/07/apple-to-congress-nothing-found-chip-hack/
Google's Titan (for servers ) chip outlined at this year's Hot Chips 2018 https://www.anandtech.com/show/13248/hot-chips-2018-google-titan-live-blog-6pm-pt-1am-utc )
A pluggable GPU is a rather poor vector for a disrupting the core "root of trust" of the system as opposed to the other aspects directly under the T2 control: boot firmware, low level power management , default boot OS image , user authentication ( fingerprint reader and/or video/audio (face/voice ) ).
The content the GPU handles is displayed on the screen anyway. It is an output to the user ; not from the user. So it is unlikely there is any user metrics to clandestine capture and anybody looking at the screen can see the output. As long as the boot drivers for the GPU are signed and authenticated, it is a highly benign subsystem device from a security standpoint.
The T2 kill switch is supposedly kicks in ( https://www.macrumors.com/2018/10/06/ifixit-repairs-2018-mbp-without-apple-diagnostics/ ) if replace the NAND daughtercards of the SSD (iMac Pro ) or cameras/logic board (hence T2 itself ) / TouchID (fingerprint ) systems.
Likewise, a "data' (non boot ) drive outside the boot process isn't all that high of a security "root of trust" problem. [ At least when not being used as a boot drive. ] . Bluetooth/Wifi is on a slippery slope. I imagine there are some ultra secure folks would would prefer if those could be unplugged from the system ( so not a reset trigger. )
Could Apple make it an optional setting. It is already in the "optional" mode now (given iFixit made replacements ). "Planned obsolesce" isn't the primary driver here really ( it has some side effect issues ) but whether this is a manufactured secure device is a problem that requires solutions.
[doublepost=1539020045][/doublepost]
Long ago SanForce SSD controllers encrypted everything by default. The data was always managed at rest compressed and encrypted. That has helpful attributes when dealing with error corrections and encodings on 'raw' NAND flash cells which can loose bits. For example several sectors of '1111' in a row means have lots of electrons to trap in a narrow area. Encrypted it can't be that uniform. ( encryptions raise the 'entropy' (randomness) of the data while compression can move the opposite way. ).
If the data is going to be encrypted anyway then can put a formal password on it. (for controllers like SanForce the nominal mode was to have a key that was built into the drive by default and just decrypt with no password 'covering' the the key. )
Apple's stance of "FileValut" being completely detached from the disk level had problems if trying to solve the issues with your own home grown SSD controller. Some of this is the mix of Apple and them being in the SSD business but only for their own devices. So their SSDs only work for their devices.
That is OK. That only becomes a substantive problem when their SSDs are the only options. ( same for GPUs implementations being the only option.). If allowed some 3rd party SSDs some of them will be encrypt/decrypting too.
[doublepost=1539020958][/doublepost]
For greater than two GPUs eGPU would likely be the default option. The MBP 15" has two GPUs but if want more than two in your set-up then it is eGPU. Mac Pro, even with less physical volume constraints, probably will fall into the same boat.
The primarily GPU being outside the box makes no sense. It leaves the Mac with a basically incomplete system (minus act of attaching a monitor or running headless ).
A second GPU ( a compute , accelerator , etc. ) would not be a huge issue if go back to desk-side like volume constraints.
Most of the GPUs on Apple's "many of these will work" list are GPUs that are part of Apple custom designs. I don't think that is an off the shelf indicator. The rest of the Mac line up also has several discrete GPUs and most of those will probably continue to "happen to work" as secondary GPU cards in eGPU or in an empty slot of a new Mac Pro. The bulk of the GPU driver work has to be done anyway.
Apple's custom embedded cards could be a derivative of the baseline reference designs. What the Mac Pro would need is "boot" and clean integration with the Thunderbolt subsystem for the primary card.
Apple ships systems with Vega in them now (iMac Pro) . They should be interested in non boot phase issues popping up with Vega anyway.
It would be helpful for Apple compartmentalizing the non boot driver development for the Mac ecosystem.
A new Mac Pro at AMD or Nvidia could spend time bootstrapping their own reference card without revealing more of new Mac system than necessary.
With that approach Apple would be pointing users towards the Windows route from the beginning .
That's not strictly true. Retail cards have different frame buffer configurations and thermal controls than the cards that Apple ships (see: All the fan issues Vega users are having on the Mac Pro right now.)
Again, not necessarily. There has been a range of issues noticed with retail cards that are not present on Apple's, that Apple seems to be collecting data on.
Maybe they're doing it to cover for eGPU as well.
Is there anything Apple has done to point to them not doing that? Everything is increasingly locked down and the T2 security chip will likely further lock down anything within apples ‘box’. I’d love to be proven wrong but everything Apple has done over the past several years points towards a further lock down-perhaps easier for Apple to swap out newer chips as they become available but not for the end buyer-or the T2 will be leveraged in the future that macOS requires it ending hackintosh’s eilther way I think if you want anything close to what most people here want (myself included) it’ll be a windows machine.
That framebuffer variance is in the eGPU cards too. I think looking too narrowly at this as a Mac Pro issue when it is really off-the-shelf versus embedded issue. The embedded versions have 2-6 display port configurations but fewer of them have something like HDMI and/or DVI that the off-the-shelf cards will have.
The long term growth of eGPU means that mismatched framebuffers is an issue that Apple will have to account for to some extent going forward. ( pragmatically there are going to be far more Macs with 3rd party GPU cards than there ever where in the 'classic' 2006-2010 Mac Pro era. ). So Apple does need to explore the edges around the expanded space they are in.
Apple has picked a handful of cards to start with ( "official" as in explicitly named and probably chased down all the bugs ahead of time), but they'll need to gradually expand that list a bit ( not most cards, but some amount of more cards ) over time.
Explore the edge probably can be used to explore what they do not want to do also. ( e.g., need to set up bug feature classifications of what going to fix and what not going to fix ) . It is big leap to spin this as Apple moving to punt completely to the 3rd party market all together.
Once past the boot phase and thunderbolt connection established what is the huge difference between a GPU put into an empty PCI-e slot inside the Mac Pro or inside the Thunderbolt peripheral box. it is a GPU card in a PCI-e card slot. The bandwidth to the Mac Pro ( v2.0 only) and ( thunderbolt x4 v3.0 ) are both at variance to nominal x16 PCI-e v3.0. Fans running out of control in a eGPU box is about as bad as fans running out of control in a MP 2010 box.
Just to add to this... Apple adding third party NVMe to the Mac Pro 5,1 is also very strange, and this time has no connection whatsoever to eGPU.
https://www.monsterlabo.com/product/the-first-batch2/
Add in 8C/16T 65W Core i7/Core i9, RTX 2070, and you get perfect Mac Pro replacement that will not fail, will not need complicated service once in a while, and you be able to use any parts you need.
Oh, and you know what? If you will buy any passive PSU - you will get dead silent computer, much quieter than any Apple computer.
Add in 8C/16T 65W Core i7/Core i9, RTX 2070, and you get perfect Mac Pro replacement that will not fail, will not need complicated service once in a while, and you be able to use any parts you need.
Oh, and you know what? If you will buy any passive PSU - you will get dead silent computer, much quieter than any Apple computer.
hi there!
Getting more and more experienced with Pcie switching (i own two xserve 3.1 and 1 maxed out 4.1 with cyclone microsytem 16x pcie gen2 5 and 10 16x slot expender) I have to say that professionaly everything revolve around Pcie. gpu/raid/in-out/network/ssd you name it , it is pcie.
even when it is a thunderbolt dongle or box, it is in fact pcie over thunderbolt.
so trully Apple could go with everything outside via thunderbolt, but they know they will never reach the performance of a cheap pc with two gen 3 pcie slot.
it would take 8 thunderbolt link and 4 calble to get the bandwith of a 16 lane pcie slot.
to exit all the lanes of a modern high end CPU they would need somthing like 10 or 12 thunderbolt link and so on.
there is simply no point doing that, otherwise you just go for an imac pro.
now... knowing Apple they could totaly come up with another proprietary pcie 16x connector, and a stack up design for each apliance... you want a gpu : you have to buy a box, you want to put a i/o card? buy a proprietary pcie expender.
but I hardly believe that would be a viable option for them on an economical stand point. to be usable each module would need it’s own 500w psu, cooling, connector with hot plug capability, it would be super hard to prevent chinese brand to buitl cheap expender clone based of what is just a pcie pass-thru connector, etc...
building a true servergrade chassis with very accessible and upgradable part is what people want, and they just need to lock it via sofrware to make customers have to buy compatible stuff.
they will make way more money selling « aproved or certified hardware » at a premium price, and no one in the real professional world will try to lowball with a « flashed PC GPU » if they can buy a certified gpu that is 30% more expensive. puting a different rom on a gpu doesnt cost as much as doing specially and proprietary gpu.
they know they have ****ed up with the new mac pro trash can, and there would be no reason to keep going this direction.
there are industry standards such as PCIE slots, and the cooling flow from front to back, and 19 » rackmount size in 1/2/3/4/5 U.
no matter how much apple executives whant to say with their own « closed system », pros who use the machine are mostly pragmatic, and if they work for a large firm, the IT buyer wont let them buy another stupid « prosumer machine » that doesnt fit in the racks.
they will say F-U either you will have a imacpro on your desk, or either you will have a 10k$ true servergrade pc/linux box in the rack with OSX runing on a virtual machine.
but recent support for NVME boot and Pcie speed fix for mac pro 5.1, i am confident that they are relying on pcie architecture for new machine, and PCIE is an industry standard that can not be altered form factor wise...
a pro machine HAVE to be able to recieve a full lengt, full height, and double with pcie card, and at least 2 of them. with all new processor having 40 lane or more, the way i see it is 4 double width slot with dynamic lane attribution just like any high end PC. that would be the most flexible hardware machine.
ad 4 thunderbolt port a couple of pcie slot and a 1000w psu and you are golden... they could go with soldered ram, if the price is just 30% more expensive it wont stop people buying it.
I dont think they have to go with bi-cpu design as now 99% of us would be happy with a 12-18 core single cpu.
so lets see what happen we are getting closer from the announcement anyway...
Getting more and more experienced with Pcie switching (i own two xserve 3.1 and 1 maxed out 4.1 with cyclone microsytem 16x pcie gen2 5 and 10 16x slot expender) I have to say that professionaly everything revolve around Pcie. gpu/raid/in-out/network/ssd you name it , it is pcie.
even when it is a thunderbolt dongle or box, it is in fact pcie over thunderbolt.
so trully Apple could go with everything outside via thunderbolt, but they know they will never reach the performance of a cheap pc with two gen 3 pcie slot.
it would take 8 thunderbolt link and 4 calble to get the bandwith of a 16 lane pcie slot.
to exit all the lanes of a modern high end CPU they would need somthing like 10 or 12 thunderbolt link and so on.
there is simply no point doing that, otherwise you just go for an imac pro.
now... knowing Apple they could totaly come up with another proprietary pcie 16x connector, and a stack up design for each apliance... you want a gpu : you have to buy a box, you want to put a i/o card? buy a proprietary pcie expender.
but I hardly believe that would be a viable option for them on an economical stand point. to be usable each module would need it’s own 500w psu, cooling, connector with hot plug capability, it would be super hard to prevent chinese brand to buitl cheap expender clone based of what is just a pcie pass-thru connector, etc...
building a true servergrade chassis with very accessible and upgradable part is what people want, and they just need to lock it via sofrware to make customers have to buy compatible stuff.
they will make way more money selling « aproved or certified hardware » at a premium price, and no one in the real professional world will try to lowball with a « flashed PC GPU » if they can buy a certified gpu that is 30% more expensive. puting a different rom on a gpu doesnt cost as much as doing specially and proprietary gpu.
they know they have ****ed up with the new mac pro trash can, and there would be no reason to keep going this direction.
there are industry standards such as PCIE slots, and the cooling flow from front to back, and 19 » rackmount size in 1/2/3/4/5 U.
no matter how much apple executives whant to say with their own « closed system », pros who use the machine are mostly pragmatic, and if they work for a large firm, the IT buyer wont let them buy another stupid « prosumer machine » that doesnt fit in the racks.
they will say F-U either you will have a imacpro on your desk, or either you will have a 10k$ true servergrade pc/linux box in the rack with OSX runing on a virtual machine.
but recent support for NVME boot and Pcie speed fix for mac pro 5.1, i am confident that they are relying on pcie architecture for new machine, and PCIE is an industry standard that can not be altered form factor wise...
a pro machine HAVE to be able to recieve a full lengt, full height, and double with pcie card, and at least 2 of them. with all new processor having 40 lane or more, the way i see it is 4 double width slot with dynamic lane attribution just like any high end PC. that would be the most flexible hardware machine.
ad 4 thunderbolt port a couple of pcie slot and a 1000w psu and you are golden... they could go with soldered ram, if the price is just 30% more expensive it wont stop people buying it.
I dont think they have to go with bi-cpu design as now 99% of us would be happy with a 12-18 core single cpu.
so lets see what happen we are getting closer from the announcement anyway...
Would love to hear some more about your experiences with PCIe expansion. Do you have any active threads (didn't see any here).