Wild speculation on new Modular Mac Pro

[Supermacguy]

I've been thinking about what Apple might do with the MMP. Despite what we all seem to want (a new cheese grater). Apple will never do that, they want to sell us the next generation.

What if the MMP had:
• 4 slots for NVME based drives (takes up less space than a single box SSD).
• 4 PCI slots, because graphics cards are still big
• 4-8 RAM slots, maybe on a daughter card like the cMP
• 4-8 slots for CPU's. This is the big change. Why this? Rumor has been around that Apple is getting iOS apps to run on desktop. And that the iPhoneX has as much processing power as a MacBook Air. And Apple's always been interested in doing everything themselves, including CPU chips.

This would make loads of sense (at least to me or us). They can ship a single CPU box with a single ARM chip. Then they can sell us daughter cards of more ARM chips. The cooling needed for these is minimal, so designing a thermal unit for each could be easy. If you wanted multiple CPUs, buy 4 (or maybe 8, or 2 per card) and voilá you've got an 8 CPU "Mac Pro".

With the experiences behind them of PowerPC, and Intel now not being able to supply what they need, they have the manufacturing capability to pump out loads of ARM chips for iPhones and then desktops. The OSes become somewhat unified, at least running on the same hardware. Then Apple is free to do "whatever they want" as they no longer are constrained by CPU design or outside supplier limitations.

 

[redheeler]

This would make loads of sense (at least to me or us). They can ship a single CPU box with a single ARM chip. Then they can sell us daughter cards of more ARM chips. The cooling needed for these is minimal, so designing a thermal unit for each could be easy. If you wanted multiple CPUs, buy 4 (or maybe 8, or 2 per card) and voilá you've got an 8 CPU "Mac Pro".

I don't understand why so many people on this forum want Apple to go through with a full switch to ARM Ax chips. Using one as a co-processor to conceivably run the iOS Simulator or real iOS is fine, but replacing Intel entirely and using it as the main processor is going to cause problems for professionals and power users.

Gone is the true desktop-class performance which has been a big focus for Intel and PowerPC chips before it; in contrast ARM has always been favored for low power consumption mobile devices like smartphones and laptops. Gone is the ability to run Windows, or Linux, or a lot of older software without slow x86 emulation.

Apple's previous architecture changes have all been about performance. The Intel Xeon chips in the 2006 Mac Pros were so much faster that even running the PowerPC software in slow emulation didn't result in a huge impact on performance. I simply don't see that happening in ARM vs. Intel.

 

[ActionableMango]

They can ship a single CPU box with a single ARM chip.

IMHO that would be the final nail in the coffin for Mac Pro.

 

[goMac]

The cooling needed for these is minimal, so designing a thermal unit for each could be easy.

The cooling for these is minimal in a phone with a mobile grade chip that downclocks quickly after doing any significant amount of work.

The cooling for an actual workstation usable ARM chip will be much more significant. ARM servers have fans and everything, they don't just have bare, uncooled CPUs like a phone.

You could design an desktop ARM chip that would work, but it's unlikely to be significantly cooler than a Xeon.

 

[ssgbryan]

With the experiences behind them of PowerPC, and Intel now not being able to supply what they need, they have the manufacturing capability to pump out loads of ARM chips for iPhones and then desktops. The OSes become somewhat unified, at least running on the same hardware. Then Apple is free to do "whatever they want" as they no longer are constrained by CPU design or outside supplier limitations.

This would be the end of the mac.

Software houses aren't rewriting their software AGAIN for the small marketshare that is MacOS.

 

[Moonjumper]

in contrast ARM has always been favored for low power consumption mobile devices like smartphones and laptops.

ARM was originally designed for desktop computers. It gained the dominant position in markets that Intel were not competitive. An ARM chip designed for desktop would be a different beast to one designed for mobile.

This would be the end of the mac.

Software houses aren't rewriting their software AGAIN for the small marketshare that is MacOS.

Xcode doesn't work massively different for the Intel based macOS and ARM based iOS as they are both based on the same core technology. Coding wise it will be a much smaller task than PowerPC to Intel was for OS X. Plus Microsoft are introducing ARM to Windows, so ARM Office, etc. will be available.

I do see issues and challenges with the change, but Apple are highly focused on their highly successful iPhones and iPads. The nearer Macs are to their successful lines, the more likely they are to survive and even thrive. Macs haven't been getting the care and attention they need for years. Hopefully that would improve with chip design in-house and a more beneficial scenario for Apple.

 

[goMac]

Xcode doesn't work massively different for the Intel based macOS and ARM based iOS as they are both based on the same core technology. Coding wise it will be a much smaller task than PowerPC to Intel was for OS X. Plus Microsoft are introducing ARM to Windows, so ARM Office, etc. will be available.

It would be the same as PowerPC to Intel, not smaller.

Xcode works just as differently as it did from PowerPC to Intel.

The one possible head start everyone will have is that people might have already ported their optimizations to ARM processors for iOS.

Just like PowerPC to Intel, a lot of applications might just recompile right over. Ones that use older API will likely have to adjust, just like the apps that used older features of Carbon for PPC->Intel. And just like PPC->Intel, if you did any processor specific optimizations (like pro apps and games do) you'll have an steep climb.

But Xcode doesn't do anything special itself to cover up the differences between processors. Code is just as processor dependent or independent as it would be without Xcode.

(Xcode also existing for PPC->Intel as well. So that's another reason that the transition would be very similar. Intel and PowerPC were also based on the same core tools and technology.)

The ARM transition is probably happening, but I'm highly skeptical that it will come to the Mac Pro anytime soon. I think we'll likely see a split platform for a while, like Apple was prepared to do with PowerPC and Intel. And I think it's a 0% chance the next Mac Pro is ARM beyond stuff like co-processors.

 

[redheeler]

ARM was originally designed for desktop computers. It gained the dominant position in markets that Intel were not competitive.

By always I really meant for most of its life, which is true. Since the early 1990's ARM CPUs have been popular in mobile devices due to their much lower power consumption, but aren't up to par with desktop workstations where power consumption and heat are less of a concern.

I fully expect an ARM-based MacBook before 2021, but not an ARM-based desktop Mac Pro or iMac Pro.

An ARM chip designed for desktop would be a different beast to one designed for mobile.

You really expect Apple to design an ARM chip for desktops that would out-perform the latest Intel and AMD offerings? This is a company that cares so little about desktops that the latest Mac Pro is from 2013 and the latest Mac mini is from 2014.

And even if they were to accomplish that, it's doubtful that a desktop-class ARM chip would run much cooler than a desktop-class x86 chip to achieve the same level of performance, plus it doesn't solve the problem of having to run certain things under slow x86 emulation.

 

[pb69]

I've been thinking about what Apple might do with the MMP. Despite what we all seem to want (a new cheese grater). Apple will never do that, they want to sell us the next generation.

What if the MMP had:
• 4 slots for NVME based drives (takes up less space than a single box SSD).
• 4 PCI slots, because graphics cards are still big
• 4-8 RAM slots, maybe on a daughter card like the cMP
• 4-8 slots for CPU's. This is the big change. Why this? Rumor has been around that Apple is getting iOS apps to run on desktop. And that the iPhoneX has as much processing power as a MacBook Air. And Apple's always been interested in doing everything themselves, including CPU chips.

This would make loads of sense (at least to me or us). They can ship a single CPU box with a single ARM chip. Then they can sell us daughter cards of more ARM chips. The cooling needed for these is minimal, so designing a thermal unit for each could be easy. If you wanted multiple CPUs, buy 4 (or maybe 8, or 2 per card) and voilá you've got an 8 CPU "Mac Pro".

With the experiences behind them of PowerPC, and Intel now not being able to supply what they need, they have the manufacturing capability to pump out loads of ARM chips for iPhones and then desktops. The OSes become somewhat unified, at least running on the same hardware. Then Apple is free to do "whatever they want" as they no longer are constrained by CPU design or outside supplier limitations.

Forget ARM. If it would still be Intel isn't 2 NVME maximum all current Intel chips can use? Also I would like to see 4 3.5" hard drive bays for cheap storage.

 

[flyinmac]

Ok... to start with, I am generally opposed to a transition to ARM for the Mac computers. But... I will say that my mind has been somewhat opened to the possibility.

Check out what these guys have managed to do with ARM

The first video is design talk. And preproduction mock ups of the design. But the information is very intriguing. Talk about integrating nearly the entire mainboard worth of chips into one ARM chip. And the raw computing power.

The next video gets into a demonstration of an actual working server class ARM processor with a discussion of actual performance measures and power consumption against an Intel XEON. Admittedly, the XEON seems less impressive now.

The last one is also a demonstration video with various aspects discussed.

In general... I don’t think migration to ARM for Mac OS would be desirable.

I was fully opposed to it. And had a couple of particular reasons why. So I decided to check out whether ARM could address those reasons. And I was surprised.

Now... I don’t think Apple is anywhere near this particular company in terms of design engineering. I think Apple looks like a novice compared to what these guys have put together.

Apple was once that smart. But old age has settled in, and now the young whippersnappers have come up and kicked butt.

So, after considering what these guys have done, yep... I think ARM can potentially kick butt.

Do I think Apple could pull it off??? Not as well.

Now... could Apple buy them and have these smart guys do what Apple doesn’t know how to do? Absolutely. And if they do, then certainly an ARM powered Mac could kick butt.

 

[goMac]

I was fully opposed to it. And had a couple of particular reasons why. So I decided to check out whether ARM could address those reasons. And I was surprised.

The thing to understand about Cavium is they make custom hardware for specific situations. So while people have had success with their server chips for servers, or their router chips for routers, those chips are only good at being server chips or router chips. It easy to be performance efficient or power efficient when you're only boosting the parts of the CPU needed for that exact problem. But if you do something like disable their acceleration on a router and go back to normal general purpose ARM silicon, they're not really that fast.

General purpose workstation ARM is a whole different issue that no one has really resolved yet. They will someday. But I'm still not aware of any workstation ARM processor that isn't specifically custom optimized for one task.

ARM isn't magic. There isn't anything inside of ARM that makes it smaller or lower power than Intel. It's just that people have been building smaller, slower things with it, or using it for applications where you only need half of the features of a general CPU, like a router.

What Apple gets out of ARM is more control and independence from Intel's schedule. Better pricing. Maybe a better GPU core. But there isn't going to be any huge change in cooling or CPU performance from Intel.

Anandtech has a pretty good review of the ThunderX that basically says the same thing. If you throw something server oriented at it, it really shines and can outdo a Xeon. Throw something not server oriented at it, and it gets kind of meh.

For example, here is the ThunderX at a common server task, kicking butt:
https://www.anandtech.com/show/10353/investigating-cavium-thunderx-48-arm-cores/15

And here it is at a media task, not so much kicking butt, because this sort of task is less useful for a server. Also worth noting that the ThunderX consumes far more power than most of the Xeons it loses to here:
https://www.anandtech.com/show/10353/investigating-cavium-thunderx-48-arm-cores/18

As much as ARM is treated like magically more power efficient than x86, part of Anand's final take was:

The Xeon D, by comparison, offers superior performance per watt: twice as good as the ThunderX. It is clear that the ThunderX is not a good match for heavy database servers, nor for enterprise workloads where energy consumption at low load is a high priority.

Again, ARM is not magic and may be optimized worse than Xeon at the high ends. If you re-balanced something like the ThunderX to be more general purpose for all types of use, you'd probably end up with something that performs slower than the Xeon and uses more power.

One of the reasons this is not really clear cut is because Intel redid their designs quite a while ago to be like an ARM chip internally. So while we think of x86 as old and crusty, they redesigned x86 to sit on top of an ARM-like foundation long ago.

 

[AidenShaw]

Forget ARM. If it would still be Intel isn't 2 NVME maximum all current Intel chips can use? Also I would like to see 4 3.5" hard drive bays for cheap storage.

Completely not the case.

I have a couple of Intel systems with five NVMe drives, and two with twelve.

All you need for NVMe is to have PCIe lanes, and with PCIe switches you can have as many lanes as you want (or can afford).

 

[pb69]

Completely not the case.

I have a couple of Intel systems with five NVMe drives, and two with twelve.

All you need for NVMe is to have PCIe lanes, and with PCIe switches you can have as many lanes as you want (or can afford).

Kool! Thanks.

 

[joema2]

...
• 4-8 slots for CPU's...They can ship a single CPU box with a single ARM chip. Then they can sell us daughter cards of more ARM chips. The cooling needed for these is minimal, so designing a thermal unit for each could be easy. If you wanted multiple CPUs, buy 4 (or maybe 8, or 2 per card) and voilá you've got an 8 CPU "Mac Pro"...

You can't put a high-performance modern CPU on a daughter card -- no card interface has enough bandwidth. That's why CPUs are in sockets, not on cards. E.g, the LGA-3647 socket used on high-end Xeons supports six-channel memory bandwidth per socket, which is 119 gigabytes/sec. No card connector system is remotely fast enough.

The Von Neumann Bottleneck is already a serious problem in computer architecture and choking CPUs by putting them on a card makes it worse: https://en.wikipedia.org/wiki/Von_Neumann_architecture#Von_Neumann_bottleneck

The main reason low-end ARM CPUs require "minimal" cooling is their computational power is also low relative to a high-end workstation or server CPU. The cooling requirements of CPUs escalate rapidly with computational horsepower. Current high-end Cavium ARM-architecture CPUs consume a lot of power. They may have a performance-per-watt advantage over high-end x86, but they'll have to prove it, not just claim it.

 

[flyinmac]

The thing to understand about Cavium is they make custom hardware for specific situations. So while people have had success with their server chips for servers, or their router chips for routers, those chips are only good at being server chips or router chips. It easy to be performance efficient or power efficient when you're only boosting the parts of the CPU needed for that exact problem. But if you do something like disable their acceleration on a router and go back to normal general purpose ARM silicon, they're not really that fast.

General purpose workstation ARM is a whole different issue that no one has really resolved yet. They will someday. But I'm still not aware of any workstation ARM processor that isn't specifically custom optimized for one task.

ARM isn't magic. There isn't anything inside of ARM that makes it smaller or lower power than Intel. It's just that people have been building smaller, slower things with it, or using it for applications where you only need half of the features of a general CPU, like a router.

What Apple gets out of ARM is more control and independence from Intel's schedule. Better pricing. Maybe a better GPU core. But there isn't going to be any huge change in cooling or CPU performance from Intel.

Anandtech has a pretty good review of the ThunderX that basically says the same thing. If you throw something server oriented at it, it really shines and can outdo a Xeon. Throw something not server oriented at it, and it gets kind of meh.

For example, here is the ThunderX at a common server task, kicking butt:
https://www.anandtech.com/show/10353/investigating-cavium-thunderx-48-arm-cores/15

And here it is at a media task, not so much kicking butt, because this sort of task is less useful for a server. Also worth noting that the ThunderX consumes far more power than most of the Xeons it loses to here:
https://www.anandtech.com/show/10353/investigating-cavium-thunderx-48-arm-cores/18

As much as ARM is treated like magically more power efficient than x86, part of Anand's final take was:


Again, ARM is not magic and may be optimized worse than Xeon at the high ends. If you re-balanced something like the ThunderX to be more general purpose for all types of use, you'd probably end up with something that performs slower than the Xeon and uses more power.

One of the reasons this is not really clear cut is because Intel redid their designs quite a while ago to be like an ARM chip internally. So while we think of x86 as old and crusty, they redesigned x86 to sit on top of an ARM-like foundation long ago.

An interesting take on it. Thank you for the details.

 

[Supermacguy]

Very good discussion! Thanks for the insights.

 

[ssgbryan]

ARM was originally designed for desktop computers. It gained the dominant position in markets that Intel were not competitive. An ARM chip designed for desktop would be a different beast to one designed for mobile.



Xcode doesn't work massively different for the Intel based macOS and ARM based iOS as they are both based on the same core technology. Coding wise it will be a much smaller task than PowerPC to Intel was for OS X. Plus Microsoft are introducing ARM to Windows, so ARM Office, etc. will be available.

I do see issues and challenges with the change, but Apple are highly focused on their highly successful iPhones and iPads. The nearer Macs are to their successful lines, the more likely they are to survive and even thrive. Macs haven't been getting the care and attention they need for years. Hopefully that would improve with chip design in-house and a more beneficial scenario for Apple.

Were you around for the PowerPC to Intel transition? It is pretty obvious that you haven't ever done this before.

Let me dumb it down for you........

Every software house has an in-house cycle (most run 18 months to 2 years) for their products. This is an unprogrammed disruption to that cycle. If Apple released an ARM Mac on 1 Jan 2019, we would be looking at a version 1 native ARM arriving on 1 Jul 2020 (at the earliest - this assumes that the product cycle also starts on 1 January). This also assumes that the software in question is selling enough copies to justify jumping architectures. Go look at how many products never made the jump from PowerPC to Intel. There is a reason that 10.6.8 hung on as long as it did.

That 1st version will be a straight port with absolutely no new features.

The 2nd version will be the 1st trustworthy version - how many software houses are going to run that risk for 10% market share?

 

[bsbeamer]

Anyone who remembers the early days of digital video editing and used legitimate Macromedia software before the simpler setups we have today will probably remember Adobe dumping Premiere on Mac and it being slightly tied to the Intel switch (after originally starting the software for Mac only).

Adobe basically used the time to merge the two builds, skipped 2+ product cycles on the Mac side, and created a much more robust piece of software that (eventually) overtook FCP7's market when Apple rebuilt with the initial FCPX versions.

That's just one piece of software from one manufacturer that took more than two product cycles to re-write for the new platform. Want to piss developers (and end users off) quickly? Make this kind of change without giving it years of ample notice. You'll end up with people flocking out of macOS after they were warmly embraced with the Intel switch, bootcamp, and machines that could run both Mac and Windows.

 

[goMac]

Anyone who remembers the early days of digital video editing and used legitimate Macromedia software before the simpler setups we have today will probably remember Adobe dumping Premiere on Mac and it being slightly tied to the Intel switch (after originally starting the software for Mac only).

Adobe basically used the time to merge the two builds, skipped 2+ product cycles on the Mac side, and created a much more robust piece of software that (eventually) overtook FCP7's market when Apple rebuilt with the initial FCPX versions.

That's just one piece of software from one manufacturer that took more than two product cycles to re-write for the new platform. Want to piss developers (and end users off) quickly? Make this kind of change without giving it years of ample notice. You'll end up with people flocking out of macOS after they were warmly embraced with the Intel switch, bootcamp, and machines that could run both Mac and Windows.

If Microsoft forces companies like Adobe to get on ARM for Windows, there might be a headstart there. But so far, I have heard absolutely no plans for Adobe CC on Windows ARM.

Windows ARM includes an x86 emulator so a lot of companies may just be fine with living in the emulator, especially for low end customers who are not likely to be big Photoshop users.

 

[barmann]

The OSes become somewhat unified, at least running on the same hardware. Then Apple is free to do "whatever they want" as they no longer are constrained by CPU design or outside supplier limitations.

Apple already are free do whatever they want , and they did .
That's the damn problem .

iOS and OSX unified ?
What could possibly go wrong ?
Or be worse ?

 

[flyinmac]

Apple already are free do whatever they want , and they did .
That's the damn problem .

iOS and OSX unified ?
What could possibly go wrong ?
Or be worse ?

Absolutely nothing. It’ll be Windows 8. Only this time the Applets will become dominant. Desktop versions will dwindle; Along with the concept of a computer being useful.

Apple doing what they do best. Being the last one to the game, but doing it better than the others. Even if it’s a bad idea, they’ll make sure everyone adopts it. Kind of like being convinced that chopping off your hand will increase productivity.

There’s a usefulness of being able to use iOS apps on the desktop where the desktop version doesn’t exist.

But in practice, we know this will lead to lazy developers taking the shortcut. Apple already gave the example. Look what happened to the iWork suite.

We’ll be using phone apps and losing desktop features. Get ready for everything being dumbed down. Cause why write a desktop version (even if it’s more powerful) when you can write once and compile and it’ll work on everything? Even if it means they all suck, at least you didn’t have to write two versions.

 

[Moonjumper]

Were you around for the PowerPC to Intel transition? It is pretty obvious that you haven't ever done this before.

Let me dumb it down for you........

Every software house has an in-house cycle (most run 18 months to 2 years) for their products. This is an unprogrammed disruption to that cycle. If Apple released an ARM Mac on 1 Jan 2019, we would be looking at a version 1 native ARM arriving on 1 Jul 2020 (at the earliest - this assumes that the product cycle also starts on 1 January). This also assumes that the software in question is selling enough copies to justify jumping architectures. Go look at how many products never made the jump from PowerPC to Intel. There is a reason that 10.6.8 hung on as long as it did.

That 1st version will be a straight port with absolutely no new features.

The 2nd version will be the 1st trustworthy version - how many software houses are going to run that risk for 10% market share?

I got my own first Mac just after the change, but have been using them since before. I have also been on teams through other big architecture changes (PlayStation to Dreamcast, XBox to Java/BREW mobile, Windows to PlayStation 2, PlayStation 2 to PlayStation 3 for example).

There is no need to dumb it down, and no need to be so dumb.

You would be looking at 1 Jul 2020 at the latest for many with an 18 month cycle, not earliest. Changes can arrive within the cycle (they need to, there is always new hardware to support). But many have left that mode anyway as they move to subscription.

Also for such a big change, Apple are likely to give advance notice, so that shortens the period from release to full support.

And it wouldn't be 10% as Windows is also getting ARM support. And a lot of software also has iOS, and therefore ARM, versions already. Office for example. Often not the full version, but it does mean the software houses have ARM knowledge.

Intel have got complacent because of a lack of competition. Having ARM on just one Mac (which almost certainly won't start with the Mac Pro, probably the Mini or MacBook instead) will hopefully get Intel moving again. I'd rather there wasn't a need for change, but things aren't great the way they currently are.

 

[AidenShaw]

The 2nd version will be the 1st trustworthy version - how many software houses are going to run that risk for 10% market share?

I think that for third party software, the share is much less than 10%.

Mac Pro sales are in the gutter - especially since the three amigos said that it was a big mistake. The fix is on the far horizon - but there's a lot of angst about whether the "fix" will be Cube 3.0 or an actual power machine.

I would venture that most Mac sales are laptops that are sold to people who would be fine with a Chromebook, but need the "prestige" of the (no longer glowing) Apple. These users will never consider a pro third party app.

The percentage of the actual Mac base for the third parties is probably less than 10% for most. And, if the third parties use network-based authentication, they'll no exactly what the percentage is and if it is dropping.

 

[goMac]

iOS and OSX unified ?

To be fair, every indication I have heard is that this is not the plan.

That would require putting more work into macOS than Apple wants or cares about.

As much work as it is to keep them apart, it's more work to integrate them. And the iOS people don't want to touch the Mac anyway.

macOS on ARM might strip away a lot of older stuff, but it won't be some sort of hybrid with iOS. It wouldn't surprise me if Apple has been dual compiling macOS on ARM for quite a while now, much like they did with PowerPC/Intel before that was public.

 

[AidenShaw]

It wouldn't surprise me if Apple has been dual compiling macOS on ARM for quite a while now, much like they did with PowerPC/Intel before that was public.

I would be surprised if they didn't run every build on ARM (and continue to run every build on PPC, and maybe even on SPARC).

If you want to prevent architecture-specific stuff from creeping into your codebase, regular multi-architecture builds are your best insurance.

And you absolutely cannot dismiss any problem simply because "it only happens on SPARC". Either your code is multi-platform, or it is not. (SPARC is an interesting case because it is "wrong-endian". PPC was "multi-endian", not sure if Apple used "big-endian" or "little-endian" for PPC. x64 is "little-endian".)