MP 7,1 What must Apple include in a new Mac Pro that we can’t live without?

[Boil]

If you look at the 2019 Intel Mac Pro with the external shell removed, just picture it with the two MPX PCIe slots removed (meaning eight slots worth of vertical space removed)...

This would bring it from a three fan system to a two fan system...

The visual proportions may be a little off, but the length is still needed to fit Pro Tools HDX cards within...

I can see Apple going either way, shorter variant of current chassis, or rebirth of the Cube...!

If they go with a new Cube, meaning external expansion via TB4/USB4, then do they make an Apple PCIe expansion chassis or leave it to third-parties...?

 

[ZombiePhysicist]

A lot of the this seems to be based on how Apple will do replaceable GPUs when the easiest answer is: They won't. There won't be replaceable GPU boards. And so there isn't a reason to imagine CPU linked boards when the easiest and most logical answer is there won't be removable/swappable boards.

Which you don't have to go far to imagine because this is all exactly what they did in 2013.

Can you explain why they wont? Why are PCI slots perceived as such an impossibility with apple chips. PowerPC machines had slots. So I see no reason why they couldn't provide PCI slots with apple chips.

 

[Boil]

A lot of the this seems to be based on how Apple will do replaceable GPUs when the easiest answer is: They won't. There won't be replaceable GPU boards. And so there isn't a reason to imagine CPU linked boards when the easiest and most logical answer is there won't be removable/swappable boards.

Which you don't have to go far to imagine because this is all exactly what they did in 2013.

Can you explain why they wont? Why are PCI slots perceived as such an impossibility with apple chips. PowerPC machines had slots. So I see no reason why they couldn't provide PCI slots with apple chips.

goMac never said they won't have PCIe slots, they said there would not be replaceable GPUs...

 

[goMac]

goMac never said they won't have PCIe slots, they said there would not be replaceable GPUs...

^ This.

I'm not bullish on PCIe slots either, especially not full width. But I was really talking about this really complicated GPU/CPU on multiple slots idea that makes very little sense for the market Apple has traditionally aimed at. And it's hard to imagine Apple selling very many machines in that configuration at that price. A quad M1 Max chiplet configuration is already going to be plenty fast for an initial version, and plenty expensive.

If eGPU support appears at WWDC, that's a good sign for PCIe GPU support. But I wouldn't bet on it. At some point maybe Apple will do a proper discrete GPU for a slotbox.

 

[phrehdd]

My very first Mac was a Mac Pro. It made sense to me as it was akin to PC counterpart with parts that can be upgraded.
Mine was loaded with 4 drives, upped the original RAM, blu ray RW and a couple of 3rd party cards. It was a great computer at the time.

To return to a Mac Pro (or Mini Pro)... I would consider after purchase ability to add drives whether SSD or standard hard drive(s). I see no reason not to be able to add at least 2 card slots for audio and possibly niche graphics cards. I would expect as before integrated drives, unified memory and graphics but have the ability to exploit the above in a manner superior to external accessories.

 

[Boil]

Muh beowolf clustah...!

 

[ZombiePhysicist]

goMac never said they won't have PCIe slots, they said there would not be replaceable GPUs...

Well who cares? It's fine if there is a built in GPU on the processor package. As long as you can put in another GPU in one of the slots, that's fine. Or are you suggesting you cannot add another GPU? Again, if so, why not. You can have multiple video cards with power PC with multiple slots. The built in GPU stuff on the processor package is essentially just another graphics card. You can have others.

Or if you cannot, I'd like to hear the reasons why not.

 

[JMacHack]

I was suggesting the processor on card solution as a possibility really. One has to admit it’d be a very “apple” move. And for the record I didn’t mean to imply the processors would communicate through pci-e, but through a special connection akin to the mpx slot.

The most likely outcome as others have suggested is that the processors won’t be replaceable, likewise with RAM and gpu since they’re on-die.

But I’m still convinced it would need some pci slots in some form. Graphics and video pros could be served with accelerator cards like the existing afterburner, and potentially M1 gpu cores on a die itself, but there’s no getting around the necessity of audio pros needing expansion cards from AVID for example.

As for storage, that’s probably gonna be the same drives as the current 7,1, minus the t2 chip.

And frankly, after the uncharacteristic apology that Apple issued, it’d be an absurd about-face to release another box without some form of internal expansion.

If they did, then what would be the difference between the Mac Pro and rumored high end Mac mini?

Apple bet on external expansion once, got blasted by the community, issued a public apology, and then released a slotbox with tons of internal expansion. It makes no logical sense that they’d try external expansion again.

 

[goMac]

Well who cares? It's fine if there is a built in GPU on the processor package. As long as you can put in another GPU in one of the slots, that's fine. Or are you suggesting you cannot add another GPU? Again, if so, why not. You can have multiple video cards with power PC with multiple slots. The built in GPU stuff on the processor package is essentially just another graphics card. You can have others.

Or if you cannot, I'd like to hear the reasons why not.

I've said a few things on this topic, but I'll repeat them:

• macOS on Apple Silicon does not currently support third party/external/PCIe GPUs. If this changes at WWDC ahead of a Mac Pro release, it's a good sign. But I wouldn't bet on it. Can't install a third party GPU if there are no drivers or software support.
• There's also a hardware support problem for PCIe GPUs. Apple Silicon does not support UEFI - which takes us back to the classic Mac Pro issues at the very least.
• Maybe Apple Silicon could support first party Apple discrete GPUs. But there have been no hardware designs or hints of a first party discrete Apple GPU so far. And a lot of the recommendations from Apple on how to optimize for their GPUs are linked to writing optimizations based around it being integrated. So it would be a big swerve for them that would need to be signaled ahead of time. Developers would need to get ready.

So yes, it's possible if Apple does certain things. But it's not possible now.

 

[DrEGPU]

On a unit basis, the 2013 Mac Pro outsold the previous generations. It was a technical failure in that they couldn't make the thermals work and fixing the design didn't make sense once they'd decided to move to Apple Silicon. If you want to toss around terms like "market failure", the most fair statement you might be able to assert is that entire Mac Pro business has been a small niche business but the idea that they were selling huge amounts of the old Mac Pro that suddenly cratered with the 2013 nMP was not actually what happened and is not what Apple stated in interviews nor what analyst reports confirmed. I have not looked into how the 7,1 has done but suspect that past the initial surge of sales around release has probably sold within the range of the previous models. I'd be super curious to know how much mainstream users overall have done serious expansion of their 7,1 models in that timeframe since Nvidia isn't supported on the 7,1 and the growing cost and limited supplies for components overall.

It outsold previous generations because it was available for a much longer time than previous Mac Pro's. It took 6 years for Apple to go from the 6,1 to the 7,1. During that time they kept selling them and basically it was a single company that was propping up the market for 2013 Mac Pros, buying hundreds at a time directly from Apple. The short of it, is that the compact size and TB2 ports made it easier to connect thousands together in a massive Xserve cluster.

Who Kept Buying the Mac Pro Everyone Hated?

As the 2013 Mac Pro ages out and its long-awaited replacement enters the frame, the trash-can-era Mac Pro will probably be remembered by most professional users as a bad memory. But it won’t disappear completely.

www.vice.com

 

[grouch]

I'm hoping there's a small form factor version with upgradable GPU, storage, and memory.

GPU not even a big deal, but having that second useless GPU on my 2013 mac pro is a bummer.

 

[deconstruct60]

I predict they will make intel based mac pros for at least another 5 years

probably not . Even more so if they choose to squat on the 2019 model.

1. the longer they keep a “sold as new” Intel Mac in the line up , then the longer they pragmatically need to support macOS on Intel . 10 year out from 2020 they very likely want to be done with new macOS on Intel upgrades. Maybe trickling out very limited security fixes , but mostly a zombie port.

2. Questionable whether Intel would want to sell W-3200 series at that point .The W-3175x is being dicontuned in 2022.

https://qdms.intel.com/dm/i.aspx/1E07901F-0EDC-42B2-8BA5-EC124761070D/PCN118523-00.pdf

shipped in 2018 . Superseded by W-3200 in 2019 . The 3175x wasn’t a full line up , so the 3200 could last 5 maybe 6 after superseded . w-3300 technically came in 2021 s o’clock is already running . Intel is using hefty discounts to keep W-3200 products still churned at most large volume workstation vendors . Thinner profits are more likely going to push that to 5 year window. ( Intel has better product to sell in 2023+ timeframe)


3. competitive wise by end of 2022 the MP 2019 will be dated on Intel/AMD side and the “half sized” Mac Pro will be eating up user base at the bottom. Sales will drop . And this time MacStadium ( and other Mac cloud services vendors ) isn’t going to save them selling ‘ancient’ product.

even if Apple switch to a W-3300 it probably won’t go past 2-3 years of new sales . 2022 -> 2024-5 -> ( +5) 2029-30 . Mainly , long enough for last gasp “ max size container box” buyers to buy .

that is long enough for Apple to iterate their multi die design to see if they had captured enough mac Pro user base to be happy to continue . ( and adjust software/ kernel foundation )

 

[deconstruct60]

….. You can have multiple video cards with power PC with multiple slots. The built in GPU stuff on the processor package is essentially just another graphics card. You can have others.

Or if you cannot, I'd like to hear the reasons why not.

as goMac pointed out the PowerPC era or video cards was not some pull some random video card into a Mac and it would work situation. The Physical socket is only ‘ half’ the issue.

the other software issues not pointed out are :

1. no direct Apple support for hardware booting to something besides macOS ( not that some hook can hang a hack on…. Actually a supported configuration . ) . So the Mac Pro 2019 meme of slapping a 3080/3090 GPU device into the container and just boot Windows to use isn’t an option.

Without placement of 3rd partyGPUs embedded in Mac products ther is no free exterior flow of GPU cards that would work in macOS.

Apple has announced that kext are going away .. so even unsigned kext drivers are not a long term solution that follow OS upgrade evolution.
There is no “new world” migration path from IOkit ( kept) to SystemsExtensions for thr GPU type. [ and some debate whether can put GPU driver into a IOMMU context needed for SystemExtensions ]

2. Apple firmware mods ( hack a Mac Pro 4,1 into a 5,1 by tweaking firmware ) . The modern Apple security stack does support that .
the core boot manager is largely a stripped down macOS . If loose that one-trueboot partition on internal Apple drive the device is pragmatically bricked. The scope of that pre-boot is highly likely going to be restricted to what Apple wants to support in known standard configurations .


3. Apple hypervisor foundation doesn’t do assignment of a pci-e device to a specific virtual machine. Can tap dance around the lack of macOS boot support if hand the video card to a virtual UEFI and virtual machine space. ( for now emulated graphics is the only path ).
The VM vendors have to use Apple’s framework as their foundation.


these are not impossible pieces for Apple to roll out later , but certainly pieces they are not putting effort into at the lowest and earliest boot context. They want to do Macs there…. which if there are ubiquitous Apple GPUs present … means doing doing just their own stuff .



P.S. AMD‘ first multi-die GPU presents as two logical GPUs. It is unclear whether Apple‘s effort in that space is going to do any better . If the 4 die/tile version presents as 4 GPUs then would be the mutiple GPU answer. They may be more focused on iterations to get to point can present that as one huge GPU .


Also pretty doubtful that Apple wants to go down the 16 pin power path of PCI-e v5 .

Gigabyte: PCIe Gen5 graphics cards to require either 16-pin power cable or "3x8-pin to 16-pin" adapter - VideoCardz.com

Gigabyte confirms that next-gen GPUs will require a new 16-pin power cable or an adapter to comply with PCIe Gen5 specs We have been covering PCIe Gen5 power connectors for a while now. The first information emerged back in October last year when rumors on GeForce RTX 3090 Ti possible specs have...

videocardz.com

if Perf/watt is the holy grail , then Apple is on a different path.

 

[ZombiePhysicist]

I've said a few things on this topic, but I'll repeat them:

• macOS on Apple Silicon does not currently support third party/external/PCIe GPUs. If this changes at WWDC ahead of a Mac Pro release, it's a good sign. But I wouldn't bet on it. Can't install a third party GPU if there are no drivers or software support.

So what. It doesnt need to. If they choose to put in PCI, obviously, this would be addressed. There is no physical or fundamental problem with doing so other than just writing some drivers/support for it.

• There's also a hardware support problem for PCIe GPUs. Apple Silicon does not support UEFI - which takes us back to the classic Mac Pro issues at the very least.

They added this before with the 2019, they can add it with a new system. Again, there is nothing stopping them other than writing driver support.

• Maybe Apple Silicon could support first party Apple discrete GPUs. But there have been no hardware designs or hints of a first party discrete Apple GPU so far. And a lot of the recommendations from Apple on how to optimize for their GPUs are linked to writing optimizations based around it being integrated. So it would be a big swerve for them that would need to be signaled ahead of time. Developers would need to get ready.

So yes, it's possible if Apple does certain things. But it's not possible now.

I don't see why they dont support a reference design like they have been doing with their GPUs for the 2019.

So it seems to me you're more pontificating on apple's WILL to do these things than on any technical barrier. If so, that is more than fair enough, and you may even well be right. But there is absolutely no technical barrier preventing them to do so, if they have the will to do so, that I've seen espoused here.

 

[goMac]

P.S. AMD‘ first multi-die GPU presents as two logical GPUs. It is unclear whether Apple‘s effort in that space is going to do any better . If the 4 die/tile version presents as 4 GPUs then would be the mutiple GPU answer. They may be more focused on iterations to get to point can present that as one huge GPU .

It would be interesting if Apple did an architecture where they did discrete GPUs that could be fused together into one, like they've been doing with Infinity Link on the current Mac Pro. But there is still the software issue - They've done a lot of tuning around unified memory and not having VRAM or local cache on a GPU. So any sort of return to big banks or caches of local GPU memory would need to be co-ordinated.

They could maybe go down a UMA path with discrete cards like AMD is trying to do. That way they could paper over the software issues. But that would be a pretty big jump we haven't seen any indication of. Feels more like a jump they could do on a 2024 slot box. Big jump from where they are now.

 

[ZombiePhysicist]

as goMac pointed out the PowerPC era or video cards was not some pull some random video card into a Mac and it would work situation. The Physical socket is only ‘ half’ the issue.

the other software issues not pointed out are :

1. no direct Apple support for hardware booting to something besides macOS ( not that some hook can hang a hack on…. Actually a supported configuration . ) . So the Mac Pro 2019 meme of slapping a 3080/3090 GPU device into the container and just boot Windows to use isn’t an option.

Without placement of 3rd partyGPUs embedded in Mac products ther is no free exterior flow of GPU cards that would work in macOS.

Apple has announced that kext are going away .. so even unsigned kext drivers are not a long term solution that follow OS upgrade evolution.
There is no “new world” migration path from IOkit ( kept) to SystemsExtensions for thr GPU type. [ and some debate whether can put GPU driver into a IOMMU context needed for SystemExtensions ]

2. Apple firmware mods ( hack a Mac Pro 4,1 into a 5,1 by tweaking firmware ) . The modern Apple security stack does support that .
the core boot manager is largely a stripped down macOS . If loose that one-trueboot partition on internal Apple drive the device is pragmatically bricked. The scope of that pre-boot is highly likely going to be restricted to what Apple wants to support in known standard configurations .


3. Apple hypervisor foundation doesn’t do assignment of a pci-e device to a specific virtual machine. Can tap dance around the lack of macOS boot support if hand the video card to a virtual UEFI and virtual machine space. ( for now emulated graphics is the only path ).
The VM vendors have to use Apple’s framework as their foundation.


these are not impossible pieces for Apple to roll out later , but certainly pieces they are not putting effort into at the lowest and earliest boot context. They want to do Macs there…. which if there are ubiquitous Apple GPUs present … means doing doing just their own stuff .



P.S. AMD‘ first multi-die GPU presents as two logical GPUs. It is unclear whether Apple‘s effort in that space is going to do any better . If the 4 die/tile version presents as 4 GPUs then would be the mutiple GPU answer. They may be more focused on iterations to get to point can present that as one huge GPU .


Also pretty doubtful that Apple wants to go down the 16 pin power path of PCI-e v5 .

Gigabyte: PCIe Gen5 graphics cards to require either 16-pin power cable or "3x8-pin to 16-pin" adapter - VideoCardz.com

Gigabyte confirms that next-gen GPUs will require a new 16-pin power cable or an adapter to comply with PCIe Gen5 specs We have been covering PCIe Gen5 power connectors for a while now. The first information emerged back in October last year when rumors on GeForce RTX 3090 Ti possible specs have...

videocardz.com

if Perf/watt is the holy grail , then Apple is on a different path.

All the above boils down to the don't currently have drivers to do this. So what. It's a new system. It's reasonable they will write new drivers. They didn't have an emulator ready to go to convert intel to appleSilicon code either. They made that happen. Nothing prevents them from writing drivers/firmware/software from making slots and 3rd party PC stuff work. There is no technical barrier at all espoused in these threads. The only speculation here is to Apple's will to make it happen. Fair enough, but that's a very different issue than some technical barrier.

Also, performance per watt is not a top priority for a Mac Pro.

 

[goMac]

So what. It doesnt need to. If they choose to put in PCI, obviously, this would be addressed. There is no physical or fundamental problem with doing so other than just writing some drivers/support for it.


They added this before with the 2019, they can add it with a new system. Again, there is nothing stopping them other than writing driver support.


I don't see why they dont support a reference design like they have been doing with their GPUs for the 2019.

So it seems to me you're more pontificating on apple's WILL to do these things than on any technical barrier. If so, that is more than fair enough, and you may even well be right. But there is absolutely no technical barrier preventing them to do so, if they have the will to do so, that I've seen espoused here.

I don't disagree with unlimited time and unlimited money Apple could do these things if they wanted (which is it's own question.)

The problem is really timing. We know this is going to be based on existing M1 Max chips. We know where the software is. We know where the firmware is. Can they make these jumps for a Mac Pro that is likely already in DVT? Highly doubtful.

But that's ok. That would fit with the rumor of a mini Mac Pro and an Intel Mac Pro both being available at the same time. If time is the issue, that buys them more time.

Again, WWDC will be a leading indicator. We'll be able to see where the software is going in advance of a new Mac Pro.

 

[deconstruct60]

^ This.

I'm not bullish on PCIe slots either, especially not full width. But I was really talking about this really complicated GPU/CPU on multiple slots idea that makes very little sense for the market Apple has traditionally aimed at. And it's hard to imagine Apple selling very many machines in that configuration at that price. A quad M1 Max chiplet configuration is already going to be plenty fast for an initial version, and plenty expensive.

Width of cards or Mac Pro ?

“….

Dimensions​

• Height: 20.8 inches (52.9 cm)
• Width: 17.7 inches (45.0 cm)
• Depth: 8.58 inches (21.8 cm)
• Height with optional wheels: 21.9 inches (55.7 cm)

…”

Mac Pro (2019) - Technical Specifications - Apple Support

support.apple.com

That would be PCI-e card length. They’d be on slipppery slope there . There are quite a few “ more than half length but short of full ; 3/4-ish” cards . But is a cube is a holy grail they are chasing perhaps willing to take hit.

Triple and quad standard width PCI-e slot . Yes. Typicallly that is driven by card having high thermal drama. Apple is likely done with that. But in the AV space there are quirky cards with multiple breakout plugs/sockets that go past the limited classic single slot width.

If eGPU support appears at WWDC, that's a good sign for PCIe GPU support. But I wouldn't bet on it. At some point maybe Apple will do a proper discrete GPU for a slotbox.

Doing a non-GUI GPGPU driver for third parties would make more sense than Apple forking the Apple GPU , Metal driver space between unified and discrete ( or at least heavy NUMA ) memory. The Apple iGPUs are next exceptionally weak. If the external card is more for computational horsepower then don’t need the whole graphics stack . ( although Apple really hasn’t done the work here ).

With fab process shrinks Apple can cherry pick off lots of workloads in the AV en/decode , AI/ML/NPU , matrix math , image , and DSP space with more fixed function logic at even better Perf/watt than GPGPU can do. The part they are likely to miss is the either even more specific ( e.g., DPU/IPU network storage card ) or more generic embarrassingly parallel computation rather than something a future iGPU missed in the single screen + GPU space.

 

[goMac]

That would be PCI-e card length. They’d be on slipppery slope there . There are quite a few “ more than half length but short of full ; 3/4-ish” cards . But is a cube is a holy grail they are chasing perhaps willing to take hit.

Triple and quad standard width PCI-e slot . Yes. Typicallly that is driven by card having high thermal drama. Apple is likely done with that. But in the AV space there are quirky cards with multiple breakout plugs/sockets that go past the limited classic single slot width.

Both directions would be a problem.

If the case is going to be smaller, Apple could just cut the number of slots, even though PCIe cards can be quite tall.

But length... The size of the Mac Pro really is because full length PCIe cards are that long. The reason I'm not bullish on a Mac Pro Cube with PCIe slots is because of the physical space. If you make the Mac Pro smaller you can't physically fit the full length PCIe cards.

Sure, Apple could ship a Mac Pro that takes half length PCIe cards. But that would really be focused on things like storage cards, Fiber Channel... I'm not even sure about AVID cards. I'm not an audio guy but they look full length to me. Maybe 3/4 length.

Apple could also just do a custom card form factor with their own GPUs and ignore the PCIe standard. But we haven't seen any sign they have discrete GPUs yet and the software support isn't really surfaced.

Full length PCIe card support would mean keep the Mac Pro the exact same depth, maybe negotiating a very small amount off if you can remove the front fans. So it's going to be a half height Mac Pro with the same depth? That would be super weird.

Doing a non-GUI GPGPU driver for third parties would make more sense than Apple forking the Apple GPU , Metal driver space between unified and discrete ( or at least heavy NUMA ) memory. The Apple iGPUs are next exceptionally weak. If the external card is more for computational horsepower then don’t need the whole graphics stack . ( although Apple really hasn’t done the work here ).

With fab process shrinks Apple can cherry pick off lots of workloads in the AV en/decode , AI/ML/NPU , matrix math , image , and DSP space with more fixed function logic at even better Perf/watt than GPGPU can do. The part they are likely to miss is the either even more specific ( e.g., DPU/IPU network storage card ) or more generic embarrassingly parallel computation rather than something a future iGPU missed in the single screen + GPU space.

Possibly. But I still go back to: Metal on Apple Silicon has really been taken in a direction that favors integrated GPUs that share memory.

Apple could roll back out discrete card support on Apple Silicon. The API is still there, even if it doesn't align with their current guidance. That's where you'd probably see announcements at WWDC. Or maybe discrete cards as a concept is dead and the next time we see them it's some sort of Infinity Fabric-ish mesh system (which I very much doubt we'd see until a few years from now.)

 

[deconstruct60]

It would be interesting if Apple did an architecture where they did discrete GPUs that could be fused together into one, like they've been doing with Infinity Link on the current Mac Pro. But there is still the software issue - They've done a lot of tuning around unified memory and not having VRAM or local cache on a GPU. So any sort of return to big banks or caches of local GPU memory would need to be co-ordinated.

The Metal additions for Infinity Link are more about directing data to be copied between GPU+VRAM instances via a "faster that frontside bus PCI-e" link as opposed to presenting on virtual , unified GPU. When it comes to presenting results on the GPU output there is copy to the hosting GPU display output controller's space. Rendezvous and coordinating between "chunks of work" on highly parallel computational workloads it works well. On a super high frame frame task it is going to be easier to just want a bigger single GPU to do the work.

I think folks are leaping toward Apple's 2-4 die GPU being a hyper gaming, 3D visuals GPU as opposed to being a better version of a Vega II Duo or W6800X Duo. The latter case was where the Metal-Infiinity Fabric links kick in.
( the NUMA hiccups for the P/E cores is likely more tolerable and can be made more transparent. ) Depends upon just how high bandwidth Apple's inter-die communication system turns out to be. I will be surprised it if is hugely better than the one on the MI200. So probably dealing with similar issues.

Also would make sense as to why Apple pursued two generations of AMD-Infinity Duo cards if the Metal software foundation would be used on Apple GPUs also. ( could have been like Afterburner though. a deployed prototype concept to incrementally get to Apple silicon solution. )

They could maybe go down a UMA path with discrete cards like AMD is trying to do. That way they could paper over the software issues. But that would be a pretty big jump we haven't seen any indication of. Feels more like a jump they could do on a 2024 slot box. Big jump from where they are now.

That would likely be a dead end as far as Thunderbolt eGPU utility goes as likely completely on a propritary connector. And Mac Pro having proprietary secondary GPU connector .... not sure if there is going to be sufficient unit volume for Apple to chase that. ( MPX bays have dual utility slots because hefty chunk of the unility lies in non MPX modules. Plus cheaper on scale to do standard stuff on a standard connector. )

A discrete GPU only die that is only good for one Mac model (Mac Pro)? That doesn't sound like Apple component reuse.

AMD's Server CPU + CDNA "compute" GPU isn't gutting FP64. It is a size and scale of computational problems Apple isn't going to go after ( "climate modeling" , "computational biology" , etc).

 

[goMac]

The Metal additions for Infinity Link are more about directing data to be copied between GPU+VRAM instances via a "faster that frontside bus PCI-e" link as opposed to presenting on virtual , unified GPU. When it comes to presenting results on the GPU output there is copy to the hosting GPU display output controller's space. Rendezvous and coordinating between "chunks of work" on highly parallel computational workloads it works well. On a super high frame frame task it is going to be easier to just want a bigger single GPU to do the work.

I think folks are leaping toward Apple's 2-4 die GPU being a hyper gaming, 3D visuals GPU as opposed to being a better version of a Vega II Duo or W6800X Duo. The latter case was where the Metal-Infiinity Fabric links kick in.
( the NUMA hiccups for the P/E cores is likely more tolerable and can be made more transparent. ) Depends upon just how high bandwidth Apple's inter-die communication system turns out to be. I will be surprised it if is hugely better than the one on the MI200. So probably dealing with similar issues.

Also would make sense as to why Apple pursued two generations of AMD-Infinity Duo cards if the Metal software foundation would be used on Apple GPUs also. ( could have been like Afterburner though. a deployed prototype concept to incrementally get to Apple silicon solution. )

Yeah, the software support they have now in Metal would not be enough. It could be the underpinnings for a virtual device that's unified.

Apple at a lot of times seems to be looking toward a multi GPU future with the Mac Pro. The 2013 was multi GPU, and the 2019 more gently so. So while I don't think Apple's just going to start throwing M1 Maxs on boards, it seems reasonable they could do linked GPUs. Just... probably not this year. Maybe Apple will surprise me though.

That would likely be a dead end as far as Thunderbolt eGPU utility goes as likely completely on a propritary connector. And Mac Pro having proprietary secondary GPU connector .... not sure if there is going to be sufficient unit volume for Apple to chase that. ( MPX bays have dual utility slots because hefty chunk of the unility lies in non MPX modules. Plus cheaper on scale to do standard stuff on a standard connector. )

A discrete GPU only die that is only good for one Mac model (Mac Pro)? That doesn't sound like Apple component reuse.

AMD's Server CPU + CDNA "compute" GPU isn't gutting FP64. It is a size and scale of computational problems Apple isn't going to go after ( "climate modeling" , "computational biology" , etc).

I think this is the problem I see too. It's a proprietary form factor. Depending on the linkage, you might need to pair proprietary cards to a matching generation with a matching GPU host. At that point, why bother with cards? Just order your Mac Pro the way you want it. Apple's accounting department would certainly like that better.

If I was going to be talked into cards, it would be an Infinity Fabric-ish meshy thing. But the simplest answer to me just seems like no GPU cards. You get what you get.

 

[Flint Ironstag]

If this Mac (that's what they should call it) can be configured to hit 100 TFLOPs of raw compute for $4k, I'm in. Otherwise, Z8 Hackintosh. May get a mini to play with whatever neural / other built-in features I'm missing.

[edit] all this talk of backplanes and dedicated systems on a MPX remind me of NeXT days and the massive daughterboards you could add to the cube: DSP, NeXT Dimension video, etc.

 

[richinaus]

Yeah, the software support they have now in Metal would not be enough. It could be the underpinnings for a virtual device that's unified.

Apple at a lot of times seems to be looking toward a multi GPU future with the Mac Pro. The 2013 was multi GPU, and the 2019 more gently so. So while I don't think Apple's just going to start throwing M1 Maxs on boards, it seems reasonable they could do linked GPUs. Just... probably not this year. Maybe Apple will surprise me though.



I think this is the problem I see too. It's a proprietary form factor. Depending on the linkage, you might need to pair proprietary cards to a matching generation with a matching GPU host. At that point, why bother with cards? Just order your Mac Pro the way you want it. Apple's accounting department would certainly like that better.

If I was going to be talked into cards, it would be an Infinity Fabric-ish meshy thing. But the simplest answer to me just seems like no GPU cards. You get what you get.

100% agree and thats how I see it going.
Despite what people want [myself included as I like the option to update the GPU] I dont see anything else coming from Apple except a 2022 version of the trash can / cube or whatever shape.

 

[ZombiePhysicist]

I don't disagree with unlimited time and unlimited money Apple could do these things if they wanted (which is it's own question.)

The problem is really timing. We know this is going to be based on existing M1 Max chips. We know where the software is. We know where the firmware is. Can they make these jumps for a Mac Pro that is likely already in DVT? Highly doubtful.

But that's ok. That would fit with the rumor of a mini Mac Pro and an Intel Mac Pro both being available at the same time. If time is the issue, that buys them more time.

Again, WWDC will be a leading indicator. We'll be able to see where the software is going in advance of a new Mac Pro.

This is not a more difficult problem than creating Rosetta. And apple has all the money. Again, youre talking about will, not technical impediments. That said, what you propose as a "middle time" where they keep both around could be a solution if they did need more time. Plausible. Agreed, WWDC will tell us a lot, fingers crossed.

 

[deconstruct60]

All the above boils down to the don't currently have drivers to do this. So what.

It doesn't. GPU devices are not even in the System Extension framework. You are saying there is no instance of a category of device drivers.... "So what". I'm saying there is not even a category of that type of "extending" driver at all. That is much deeper "hole" to dig out of.

In modern macOS, the kernel is Apple only land. If there is no way in from the outside then there is no "exit ramp" for the 3rd drivers to get off to get to "kernel land". [ If there is freeway that goes though a town and there is no exit. then no one is getting off. There is no easy fix to that after the freeway is constructed. It is also demonstrative of intent. ]

It's a new system. It's reasonable they will write new drivers. They didn't have an emulator ready to go to convert intel to appleSilicon code either. They made that happen.

Rosetta 2 was there at launch because they put effort into it. Apple had another whole year after launch to minimally put up a GPU class of system extensions even if didn't sign any extensions implementation. They did not.
They did say that "time is almost up" for USB and basic I/O extensions to switch over at 2021 ( decent chance they'll turning off that class of kext on this or next macOS upgrade cycle. ).

Nothing prevents them from writing drivers/firmware/software from making slots and 3rd party PC stuff work. There is no technical barrier at all espoused in these threads. The only speculation here is to Apple's will to make it happen. Fair enough, but that's a very different issue than some technical barrier.

For non GPU's it already works. There are 30+ cards on the TB external PCI-e card enclosure that work with the M-series Macs:

(PDF file.)

https://www.sonnettech.com/support/downloads/manuals/TB_PCIe_Card_Compatibility.pdf

Echo III Desktop (Thunderbolt to PCIe Card Expansion System) - SONNETTECH

Dual-interface Thunderbolt expansion system with 2 PCIe slots and auxiliary power for high-bandwidth DV I/O PCIe cards. For Thunderbolt 4 and 3 computers.

www.sonnettech.com

That work got done. There are lots more that are not, but those tend to be either old (not getting much new driver work done for them) , not in the new System Extension class hierarchy, or coupled to early boot environment issues ( storage boot entanglements).

Also, performance per watt is not a top priority for a Mac Pro.

Says You. You are not Apple. The most consistent topic they drone on in every M-series presentation so far is Perf/Watt. To claim is not an Apple priority is a bit delusion. That priority isn't likely to change just for a product that is around 1% of the Mac product mix.

It is similar to the years between 2010-2019 where folks said Apple should dump Thunderbolt to make rolling a new cheesegrader easier ... and yet Apple never did that. Probably not now either. There will be Thunderbolt controllers on the die in the package this next round too. And the system is going to a priority on Perf/Watt. That is about the only way going to close pack 2-4 dies together and not take a single thread performance hit.

Like Thunderbolt is a priority with the rest of the Mac product line up leads to it becoming a "basic Mac property", Apple is doing effectively the same thing with Perf/Watt. Apple is shifting to tighter thermal enclosures for the rest of the line up. Chucking dGPUs. It is the same core design constraint is going to leak into the Mac Pro M-series packages because the die implementations are all related on the same family. Apple is extremely unlikely starting over from scratch and/or off on a different fab process for the Jade2C and Jade4C dies relative to the Jade-chop and Jade dies (M1 Pro and Max).

Apple is rumors to coming out with a "half sized". Mac Pro. Going to need to be some power saved somewhere to chuck half of the volume of the current Mac Pro ( much of that internal volume is primarily for thermal dissipation. If the volume is going then so is the thermal dissipation issue that drove the volume jump in first place. )

The Intel solutions rested upon a much bigger ecosystems than just the Macs. There was a broader subset of CPU to pick from there. It is not going to anywhere even close to near the same breadth when Apple is doing them all themselves. Even less so for the relatively very low volume products. There will be just enough adaptation to cover the difference but the basic design infrastructure is likely going to be more similar to laptops than data center servers.