2018 Modular Mac Pro might take lead from Mac Mini

[Omega Mac]

the mini will need an server class cpu to have the pci-e to do any thing like that.

The Base might not if you run with the concept for a moment.

So you buy a base knowing you can buy any other modules you like but for now you are happy with the base. It's stand alone working Mac Mini Pro... it's $1000. (it might have a server class CPU but only one maybe max 2) and one HD. There could be an option for two standard bases based on single or Dual CPU's. logic board.

Maybe then later you can add a CPU module if you like, with 4 CPU's! Now you have 6 CPU's. Think the same for GPU's, HD's and so on. From what I can see there is flexibility within the PCI-E standard to accommodate types of design.

Where this might fall down is power and thermals.

Modular if done right can be any common virtually you might want or you can add a module with HD & CPU upgrade.

You can buy into a system. The other modules need not change (just like mini has been the same case for year but with ever changing inners) but you can order varying inners empty or pre-installed whatever.

This is a bad idea.

Yet you posit not a single reason. Why bother posting a one liner it adds nothing to the sum of the discussion?

It's very simple. It's not a bad nor a good idea. It's an idea that has potential and worth exploring.

It might not be the right idea or it might be but you can see already we have three manufactures playing with in the past or more recently this form factor and this idea. I include Apple in this because I think the Mac Mini form factor inspired and lead Acer & HP.

 

[theluggage]

Yeah... the Pascal driver release happening at the same time as nVidia's own Titan Xp launch was just a coincidence, but happening around the same time as Apple's press conference is irrefutable proof that the new Mac Pro will have an upgradeable GPU.

It definitely wasn't because Apple just announced nVidia cards were going to work in shipping Apple products again.

Wow - you got something right - apart from the "in" bit - Apple have (effectively) announced that nVidia cards are going to work with shipping products - as TB3 eGPUs supported in 10.13 (and already possible with a bit of a hack, I understand) which is likely to be popular with heavy CUDA users.

You don't think that's a possible explanation (not that any explanation beyond "they just got around to it and/or were waiting for the Titan XP launch" is needed) for nVidia releasing Pascal drivers? More likely than "they needed to release Pascal drivers publicly in April so that people could download them ready for the new Mac Pros which will be delivered in a year or two?"

OK, lets assume you're right: say Apple told nVidia that the 2018 (hopefully) Mac Pro was going to support nVidia and this persuaded them not only to start Mac developing drivers again but to release them as a public beta (despite them not being interested in selling Titan XPs to Hackintoshers).

Now explain why that wouldn't apply if the future Mac Pro had a non-user-replaceable GPU fixed to the motherboard or otherwise buried beneath a custom cooling system (You know, like the one in the only pro machine that they have pre-announced in detail) with nVidia available as a BTO option for CUDA users? Or an Apple-branded TB3 external GPU option?

So, even if you're right (its not an impossible combination of events - they're just not logically required by any available evidence) its irrelevant to the original argument of whether the new MP will have a user-replaceable GPU (or anything).

Of course, it might still do - if I could reliably predict this sort of thing I'd be off to Wall Street to make billions. I'm only pointing out that nothing said in that interview transcript unambiguously states that the new Mac Pro will have either replaceable RAM, SSD or GPU - something you refuse to entertain even as a possibility.

The other possibility is that they're still at the multiple design candidates stage and are running the sealed iMac Pro idea up the mast to see how many of their pro customers salute before deciding which way to go. That, by the way, is why Apple would be carefully using non-committal language at this stage in the game.

Maybe actual working Pros mainly get their equipment on business lease, and will be happy with better BTO options? Maybe its only the tinkerers and enthusiasts here on MR who give a wet slap about aftermarket upgrades?

 

[theluggage]

Thanks for the link, I enjoyed reading the transcript. However, there are several Apple quotes from John Gruber, who was actually there, which do not appear in Techcrunch's transcript.

However, my concern stems from trying to deduce by what Apple mean by words like "upgradeable" or "modular" (because words mean precisely what the speaker wants them to mean...)

For example (my emphasis):

Craig Federighi: But certainly flexibility and our flexibility to keep it current and upgraded. We need an architecture that can deliver across a wide dynamic range of performance and that we can efficiently keep it up to date with the best technologies over years.

...so "upgrade" can mean Apple's flexibility to keep the product up to date, not necessarily the users'. So who knows what sort of "upgrade" they're talking about in other quotes? And from other quotes in the transcript, all you can say about their understanding of "modular" is that it means "not an all-in-one or laptop".

Bottom line: Apple makes more money selling you a custom BTO option, to the specification you'll need for the next 3 years, than selling you a base system that you can upgrade "as and when" with third party components. If they can keep to that business model without decimating their customer base (but they can probably afford to write off a few), they will.

An awful lot of true pro customers (i.e. those that can make a business case for a $10 grand workstation) renew their systems on a 3 year cycle for tax efficiency reasons (with extended warranties or via a leasing+service scheme) and won't care so much about mid-life upgrades as the enthusiasts here on MR.

 

[Joe The Dragon]

if you watch the talk show live wwdc episode, Craig specifically states that a gpu on an eGPU case has a higher latency than a bus direct slot, so Apple is aware that a PCI slot inside a computer has capability advantages over a PCI slot on a thunderbolt bus. How much weight they give that is open to question.

also speed is a lot slower pci-e X4 max less if system DP data is also running over the same bus.
[doublepost=1497405274][/doublepost]

Or a PCIe switch with the quite reasonable caveat that if you try to run all devices at full speed at the same time, there will be some throttling.

Why do so many people have the perception that the number of device PCIe lanes has to exactly match the number of CPU (+PCH) lanes? Very few workloads need everything to run a peak bandwidth simultaneously....

so you want to run 1-2 pci-e SSD's + an video card stacked off an pci-e X4 DMI link?
[doublepost=1497405512][/doublepost]

The Base might not if you run with the concept for a moment.

So you buy a base knowing you can buy any other modules you like but for now you are happy with the base. It's stand alone working Mac Mini Pro... it's $1000. (it might have a server class CPU but only one maybe max 2) and one HD. There could be an option for two standard bases based on single or Dual CPU's. logic board.

Maybe then later you can add a CPU module if you like, with 4 CPU's! Now you have 6 CPU's. Think the same for GPU's, HD's and so on. From what I can see there is flexibility within the PCI-E standard to accommodate types of design.

maybe just maybe with AMD cpu-s and it may be to far of an io run. AMD has flex link that can drive some pci-e lanes as CPU to CPU linking but X4 TB3 is why to slow and the TB controller over head lag may kill it.

 

[AidenShaw]

so you want to run 1-2 pci-e SSD's + an video card stacked off an pci-e X4 DMI link?

No, I'd run 24 PCIe 3.0 lanes into a switch with 40 PCIe 3.0 lanes out.

 

[Omega Mac]

also speed is a lot slower pci-e X4 max less if system DP data is also running over the same bus.
[doublepost=1497405274][/doublepost]
so you want to run 1-2 pci-e SSD's + an video card stacked off an pci-e X4 DMI link?
[doublepost=1497405512][/doublepost]
maybe just maybe with AMD cpu-s and it may be to far of an io run. AMD has flex link that can drive some pci-e lanes as CPU to CPU linking but X4 TB3 is why to slow and the TB controller over head lag may kill it.

Why do you think the reach would be to long? It might not have to be.

The mac mini case is 3.6 cm in height. You have four sides to use per mac min pro module.

What if the logic board is also modular in nature. Click and connect they become one.

So I'm not suggesting TB3 for linking.

Curiously, what footprint on a board is the smallest intel processor (m?) versus a server grade or i5/7/9? Forgetting things like heatsinks or what not for the sake of the hypothetical.