MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)

[goMac]

Matthew Panzarino (TechCrunch): What’s your philosophy on external GPUs?

Craig Federighi: I think they have a place.

Matthew Panzarino (TechCrunch): Seems like it would have offered the maximum flexibility in the space where you would never have to worry about thermal problems theoretically as long as the external GPU was built right.

John Ternus: I think there’s some aspects of them where they’re going to be beneficial and there’s some workloads where they’re going to be less effective.

https://techcrunch.com/2017/04/06/t...-john-ternus-on-the-state-of-apples-pro-macs/

Sure. I think what they’re saying is that with a 4x, physically longer link there are going to be some issues.

What I’m saying is I could see Apple spending the last three years fixing that with a new short distance link.

And in Apple’s mind a proprietary link that could carry 8x or 16x traffic would be equivalent to an internal PCIe slot. And if each module had it’s own cooling, that “solves” the cooling problem they had with the 2013.

It’s a fever dream that sounds way too much like Apple. And it also explains why it’s taking three to four years. A backplane concept would have taken less time.

 

[Boil]

I would personally hope for the backplane idea over stackable. But again, stackable sounds like exactly the sort of thing Ive would come up with. Backplane isn't "Apple" enough.

Then the question is, discrete CPU & GPU(s), or a plethora of ARM APUs...?!?

 

[jscipione]

I am guessing that Apple's philosophy is to use 1 internal GPU + n external GPUs based on their presentation from WWDC 2018. I could be wrong but I don't think Apple is capable of coming up with a speedier bus than TB3 for the upcoming Mac Pro. They rely on Intel to provide their I/O which means TB3 for now and USB4 on Arm sooner or later, which is essentially the same thing.

 

[danwells]

They'll almost certainly have at least one internal GPU on every configuration - I don't think they're that dumb (if nothing else, they can't outperform the iMac Pro without one, and they've promised to do that).

I'd give them (pretty much a guess) a better than 50/50 chance of supporting two internal GPUs (probably not required in every configuration - they got enough flak from audio folks for that on the trashcan).

I'd give them a decent chance of having some external connection faster than TB3 to make high-end eGPUs more attractive for people who want more than two. This could be as simple as a bunch of TB3 buses with some way of bridging two or even four

 

[askunk]

Here we are. The LG 5K display joins the 4K in leaving the Apple Store.

The new Pro display is coming. What's the point of presenting a 6K display if you don't have a Pro machine that can use it? Does anyone think that Apple could come out at WWDC with the TBD without contextually announcing the mMP?

 

[mattspace]

so... the 7,1 has 2 triplewide pci slots, each with dual 8 pin power, and is a machine designed & engineered specifically for hypervising. You can put an Nvidia GPU in one slot, and AMD in the other, and the T(x) chip handles the bare metal and hardware passthrough to the dual concurrent OSes so you can have any 2 of Windows, Linux & a macOS (which will only use the AMD card) running.

The new bare-metal-running operating system/kernel is called " PrOS" ("Apple Pro-ess" pronounced with Jobsian weirdness to challenge jag-wire as "prowess").

Anyway, back to grim reality.

 

[ETN3]

They'll almost certainly have at least one internal GPU on every configuration - I don't think they're that dumb (if nothing else, they can't outperform the iMac Pro without one, and they've promised to do that).

I'd give them (pretty much a guess) a better than 50/50 chance of supporting two internal GPUs (probably not required in every configuration - they got enough flak from audio folks for that on the trashcan).

I'd give them a decent chance of having some external connection faster than TB3 to make high-end eGPUs more attractive for people who want more than two. This could be as simple as a bunch of TB3 buses with some way of bridging two or even four

Thundebolt 4 should be around the corner...

 

[Mikael H]

plethora of ARM APUs

That sounds like a good way to switch the potential user base for the new Mac Pro from a "low single-digit percentage" of all Mac users to a "small fraction of one percent" of them. How many applications do you know, that scale well across a Beowulf-style cluster?

 

[Boil]

It’s a fever dream that sounds way too much like Apple. And it also explains why it’s taking three to four years. A backplane concept would have taken less time.

Then the question is, discrete CPU & GPU(s), or a plethora of ARM APUs...?!?

That sounds like a good way to switch the potential user base for the new Mac Pro from a "low single-digit percentage" of all Mac users to a "small fraction of one percent" of them. How many applications do you know, that scale well across a Beowulf-style cluster?

Maybe that is where the extra time went, into creating a new Rosetta Stone software to assist in the transition to ARM & software to handle the scaling issues...?

Besides, I hear the NeXT cube hosting Steve's consciousness imprint needs replaced, and my hypothetical ARM cluster modular Mac Pro cube would be perfect...!

 

[ssgbryan]

Maybe that is where the extra time went, into creating a new Rosetta Stone software to assist in the transition to ARM & software to handle the scaling issues...?

Besides, I hear the NeXT cube hosting Steve's consciousness imprint needs replaced, and my hypothetical ARM cluster modular Mac Pro cube would be perfect...!

No - just no.

I lived through that once already - it is why I stayed on 10.6.8 until 10.12 came out.

A lot of software won't get ported, and the performance in emulation will suck. It would be different if Apple had a bunch of high-speed, low drag software engineers, but that Apple is gone.

 

[deconstruct60]

Thundebolt 4 should be around the corner...

Unless you are trying to cross label USB 4 as thunderbolt 4 , no it shouldn't be just around the corner. Now that Thundebolt has been assigned to an even larger and more diverse committee of USB-IF, its development will most likely more slower now... not faster.

USB-IF has to first fniish digesting Thunderbolt 3. Then at some future point after the committee has all agreed on how to move forward ... do something. that is probably going to take a while. The natural topology of Thunderbolt networks is not the classic USB one. There are still no implementers separate from Intel. Those folks are going to want to get up and running before the foundation is shifted yet again.

The major key to Thunderbolt ecosystem development right now is adoption.; not yet another (possibly incompatible in some way physical or implementation cost structure way ) version.

 

[Blair Paulsen]

Imagine a version of the trash can cylinder that has a curved tail, think the shape of a comma from top view. Would give it that Apple design flair they understand how to sell. The daughter case would be identical in shape, but flipped (mirrored) to create a yin/yang meeting of the tail extensions- but with some airspace to limit cooling compromises.

I fully grasp why a proper tower is a better idea than any of this - but since the chances of them going back to a big box design seem remote...

Why I think they might have something along these lines in the pipeline...

If you cover a more diverse group of market segments with a broad range of BTO - by order or by retrofit - offerings, you might realize some unit cost savings. (Small numbers, different math, but still some.)

Group 1) "I need a new one" (reasons abound)
If a single piece solution with the right BTO can meet your needs then it might be a popular replacement for the tcMP in a lot of shops. TB3 may not be the ideal bus for compute tasks like eGPUs, but it can support a ton of fast storage if you give it enough headers/lanes. This means having a lot of internal storage capacity is not as important as it was when the cheese graters with the 4x 3.5" sleds came out.

Group 2) Media Creators
Need to push some pixels (samples, whatever) and need some serious GPU resources. Want a hot rod that doesn't need a full time mechanic (read IT support). Let me pack the daughterbox with whatever GPUs work best with my key software tools, or give me an AMD option that is optimized enough to make CUDA cores irrelevant. Budget challenged buyers still have TB3 for eGPU - some speed penalty vs custom daughterbox, but if Apple puts in some work on the drivers... (rim shot).

Group 3) Everything Bangers
High end. Corner cases, server rooms, media rendering, VR, etc that will always need more resources. Some Enterprise customers, media mastering/serving/distribution, etc. Yes, the bulk of this market is OS agnostic (or Linux lovers), but there are still some Mac-centric educational institutions and custom implementations that don't want to change platforms (whether they should, or not).

I'm also intrigued by the idea that the "brain" tower could be swapped down the road while retaining the daughterbox and it's high speed connection. Most of us on this forum consider swapping internal drives, RAM, etc as easy as making toast - but for a sizable group of customers who might want a high ratio of power to complexity, a single custom "plug" that allows most users to swap out brainbox or daughterbox sounds pretty good. Might support some interesting replace/repair programs to simplify servicing, etc... Think Intel's tick-tock strategy applied to two physical cases.

 

[Mago]

A more less plausible leak suggest the mMP to be much like the cheese grater but turned 90deg for vertical airflow, to use 2x PCIE GPU no extra PCIE slots, GPUs to include internal dp1.4 headers to feed the tb3 controller, only RAM & CPU upgradeable (PCIE SSD super faster (8x bus?), No word on CPU to assume Intel Xeon W and multicore Apple Axx, will be all black parts satin finish and some parts shinny black.

 

[AidenShaw]

I'm also intrigued by the idea that the "brain" tower could be swapped down the road while retaining the daughterbox and it's high speed connection.

This might sound interesting, but history shows us that new systems often come with new higher performance interconnects. Crippling a new "brain box" by using the previous generation interconnects would be embarrassing.

What if Apple had created a "daughterbox" to connect to the tcMP6,1 via T-Bolt 2.0? When the tcMP7,1 arrives, how many will be happy with a T-Bolt 2.0 daughterbox connected to the T-Bolt 3.0 tcMP7,1? (And that's making the big assumption that different generations of brainboxes and daughterboxes will even be compatible at the slowest speeds.)

This trend goes back decades (8-bit SCSI, 16-bit SCSI, Ultra SCSI, PCI 32, PCI 64, PCI-x, PCIe, PCIe 2.0, PCIe 3.0, DDR, DDR2, DDR3, DDR4, T-Bolt, T-Bolt2, T-Bolt3, ....)

 

[Peace]

Man some of these ideas are more art design than a computer..

Get with it people. It's a computer not the Sistine Chapel.

 

[ekwipt]

Thundebolt 4 should be around the corner...

will it just be USB 4 though? Which would pave the way for non Intel (think AMD ) Macs?

 

[fuchsdh]

Sure. I think what they’re saying is that with a 4x, physically longer link there are going to be some issues.

What I’m saying is I could see Apple spending the last three years fixing that with a new short distance link.

And in Apple’s mind a proprietary link that could carry 8x or 16x traffic would be equivalent to an internal PCIe slot. And if each module had it’s own cooling, that “solves” the cooling problem they had with the 2013.

It’s a fever dream that sounds way too much like Apple. And it also explains why it’s taking three to four years. A backplane concept would have taken less time.

As others have pointed out, unless these modules are truly breathtakingly simple to put together, it doesn't sound like Apple at all. "Let's have multiple redundant power supplies and fans to make more noise" is like nothing they produce. They might be fine with you having to add onto your naked robotic core with said loud eGPUs, etc., but they don't produce one themselves and never will (and hence the Blackmagic eGPU exists as basically Apple's eGPU but not produced by them.)

It's less than a month to go to WWDC, feels like it's a waste to keep spinning wheels on fanciful ideas and kremlinology when we'll likely know soon enough.

 

[Flint Ironstag]

But I am not talking about individual 'boxes / modules', I am talking about a singular cube chassis with an ultra-high-speed backplane & the actual "modules" are the daughtercards...

Nobody is asking for custom daughtercards.

 

[AidenShaw]

It's less than a month to go to WWDC, feels like it's a waste to keep spinning wheels on fanciful ideas and kremlinology when we'll likely know soon enough.

Should there be a poll on whether Apple will say anything about the vwMP at MacWorld SF 2019?

My vote would be "No mention at MWSF 2019 - but instead a minor press release on the last day of Saturnalia - followed by a ritual delivery of one system to one customer a week later".

 

[Boil]

Nobody is asking for custom daughtercards.

Cool that you speak for the entirety of the human existence...

Daughtercards in a singular chassis seems way more preferable to a collection of modules stacked up with proprietary connectors...

 

[Aldaris]

Cool that you speak for the entirety of the human existence...

Daughtercards in a singular chassis seems way more preferable to a collection of modules stacked up with proprietary connectors...

But Tim as all that stock in soldering

 

[fuchsdh]

Should there be a poll on whether Apple will say anything about the vwMP at MacWorld SF 2019?

My vote would be "No mention at MWSF 2019 - but instead a minor press release on the last day of Saturnalia - followed by a ritual delivery of one system to one customer a week later".

Considering there's no Macworld SF 2019, I'm going to bet a lot that there will be no Mac Pro news there

 

[AidenShaw]

Considering there's no Macworld SF 2019, I'm going to bet a lot that there will be no Mac Pro news there

3 June 2019 to 7 June 2019

 

[Boil]

Should there be a poll on whether Apple will say anything about the vwMP at MacWorld SF 2019?

My vote would be "No mention at MWSF 2019 - but instead a minor press release on the last day of Saturnalia - followed by a ritual delivery of one system to one customer a week later".

Considering there's no Macworld SF 2019, I'm going to bet a lot that there will be no Mac Pro news there

3 June 2019 to 7 June 2019

Your linking to a c|net article regarding WWDC 2019 does not change the fact that you referenced MacWorld SF...

 

[mattspace]

A more less plausible leak suggest the mMP to be much like the cheese grater but turned 90deg for vertical airflow, to use 2x PCIE GPU no extra PCIE slots, GPUs to include internal dp1.4 headers to feed the tb3 controller, only RAM & CPU upgradeable (PCIE SSD super faster (8x bus?), No word on CPU to assume Intel Xeon W and multicore Apple Axx, will be all black parts satin finish and some parts shinny black.

so a less-good version of what you can already do with commodity parts and Streacom's F12C?

or tipped onto its short end?

2 GPUs, wonder if they'll be stuck with the laughable "one is for compute only" configuration, or if they'll actually expose their connectors to the world.