MP 7,1 What does Apple's switch to ARM based processors mean for the future of the Mac Pro

[Macintosh IIcx]

PCIe is an industry standard and it's not tied to Intel. They don't even offer the latest version of PCIe on any of their products currently.

PCIe as a standard is supported on Intel processors and Apples own ARM processors as-well as SPARC, IBM Power, the Open RISCV processors, AMD processors etc

There would be literally no point at all for Apple to abandon it. Similarly Thunderbolt is no longer tied to Apple and third party host chips that support it are coming to market. Apple could integrate Thunderbolt into their own SoC designs going forward.

I get what you're saying regarding that Apple could go fully custom if they wanted but there really is no reason to do that with PCIe it has a very clear and concise roadmap and won't cause them any bottlenecks. The switch away from Intel solves real problems they are having by contrast.

Yeah, it will be interesting to see how this pans out. For the entry level Mac ARM devices, they will probably go with their own ARM CPU+GPU like on iPad so they wont need PCI-E here. The big question is whether they will keep AMD’s GPU in the more powerful solutions because then you’re right they will need to keep PCI, but if AMD’s GPUs are out the door also... well then everything is up in the air.

 

[Quu]

Yeah, it will be interesting to see how this pans out. For the entry level Mac ARM devices, they will probably go with their own ARM CPU+GPU like on iPad so they wont need PCI-E here. The big question is whether they will keep AMD’s GPU in the more powerful solutions because then you’re right they will need to keep PCI, but if AMD’s GPUs are out the door also... well then everything is up in the air.

Even the iPhone and iPad Pro use PCIe to talk to their built-in NVMe NAND storage.

Heck on Intel systems they have this thing called the DMI which is where the CPU talks to the Chipset but it's basically just PCIe with some protocol changes. AMD does the same for their CPU's to their Chipsets and even inside their CPU's for their own chiplets to talk to each other they use PCIe connections that run their own custom protocol called Infinity Fabric. You can think of DMI and Infinity Fabric as like custom implementations of PCIe in the same way Thunderbolt is.

 

[Macintosh IIcx]

Even the iPhone and iPad Pro use PCIe to talk to their built-in NVMe NAND storage.

Heck on Intel systems they have this thing called the DMI which is where the CPU talks to the Chipset but it's basically just PCIe with some protocol changes. AMD does the same for their CPU's to their Chipsets and even inside their CPU's for their own chiplets to talk to each other they use PCIe connections that run their own custom protocol called Infinity Fabric. You can think of DMI and Infinity Fabric as like custom implementations of PCIe in the same way Thunderbolt is.

Fair enough, but hopefully you got my point: If Apple decides to use their own custom ARM CPU and GPU, They can do a lot of custom stuff that Mean they doesn’t support plug and play with standard PC components like a GPU.

I still think they will, like supporting E-GPU over Thunderbolt (PCI-E) and Maybe even PCI-E slots on a new expandable Mac Pro, but it is not a sure thing in my book.

 

[RaoulDuke42]

Does anyone really think Apple wants to go through another 7,1 type iteration even with a change to ARM?
It's such a niche machine... and it doesn't seem to be Job's passion product.

Pretty sure dude hasn’t had any “passion projects” for about nine years now. Unless I’ve slipped into a parallel dimension where he didn’t pass on in 2011?

 

[dannys1]

I honestly don't it'll come to Mac Pro - I see more sense in them using AMD Threadripper for top end machines and ARM for portables that require less processing.

 

[Quu]

I honestly don't it'll come to Mac Pro - I see more sense in them using AMD Threadripper for top end machines and ARM for portables that require less processing.

That would mean they'll have to maintain two separate architectures on Macs in the market at the same time which to me just seems extremely unlikely.

Not only would it increase development effort for both Apple and third party developers but it would also create inconsistencies between platforms. Already today there is a wide spectrum of x86 based Intel processors being used in Macs with features some processors have and some do not.

For example some processors have integrated graphics and offer Intel QuickSync which is a video encoder/decoder that is highly efficient and desirable. It can accelerate H.264 and H.265 encoding and decoding while using very little of the normal cores present in the Intel CPU's which saves power and reduces heat while being much faster.

This feature is not present on the iMac Pro or the Mac Pro because these processors do not have integrated graphics in the CPU package and thus do not contain Intel QuickSync. Contrastingly these Macs contain processors which feature the AVX-512 instruction and that is not present on the normal iMac, MacBook or the MacBook Pro's.

This instruction extension can accelerate performance for scientific simulations, financial analytics, artificial intelligence (AI)/deep learning, 3D modeling and analysis, image and audio/video processing, cryptography, and data compression.

These are just a couple of differences I chose because they're easy to convey but if you look at the iPhone for instance it has a dedicated area of transistors for machine learning. This area of compute is growing rapidly as it becomes used for more and more things, it's not just important for facial recognition and self-driving cars.

Eventually your computer could leverage this for better compression of files, better blocking of ads on the web, better searching of files on your computer, better noise suppression on your microphone etc

Since Apple is already pressing forward with this on their A series chips it stands to reason they would want to bring the same innovations to the Mac and that would mean a full top to bottom processor rollout to keep hardware features consistent in my opinion.

This inconsistency on the Intel processors has caused Apple consternation on multiple occasions for instance when they first released Sidecar for the iPad and some Macs didn't support it due to the need for efficient H.265 video encoding. Apple eventually included such an encoder in the T2 security chip as a way to make all Macs (iMac Pro, Mac Pro etc) consistent in this feature availability, something they wouldn't need to do if they made the entire SoC themselves.

 

[high heaven]

Maybe once they able to start making better ARM chips, why not making the ARM server as they did before? The server market is quite huge and that's the main financial source where Intel is getting a lot of profits. Workstation is such a niche market but not a server since people need to pay tons of money for making a server.
[automerge]1592105459[/automerge]

I honestly don't it'll come to Mac Pro - I see more sense in them using AMD Threadripper for top end machines and ARM for portables that require less processing.

Threadripper is not comparable to Xeon but Epyc.

 

[Quu]

Maybe once they able to start making better ARM chips, why not making the ARM server as they did before? The server market is quite huge and that's the main financial source where Intel is getting a lot of profits. Workstation is such a niche market but not a server since people need to pay tons of money for making a server.

I don't think Apple sees the Mac Pro segment as a way to generate profits to be honest. I think they see it more as a halo for professionals to keep them using Apple products and have that as a way of marketing this idea to the masses that Apple is the best so buy our products that you can afford.

And because of that I don't see them entering the server market beyond continuing to make the Mac Pro available in a rackmount configuration. You could of course use these as servers but they lack any kind of remote management (IPMI) so their usefulness in a datacenter is quite limited and/or require a lot more hands-on management from physical employees making them unsuited for this kind of deployment.

Threadripper is not comparable to Xeon but Epyc.

XEON as a brand is used by Intel across their entire product stack now from small ITX based 8 core systems like the latest NUC (to be used by professionals as workstations) to large single and dual-processor workstations to servers.

Threadripper targets the HEDT (high-end DeskTop) and workstation markets similar to some of the XEON range while AMD's EPYC processors face off against the most expensive XEON workstations (1.5TB+ memory support, 64 cores, dual-processor etc) and servers.

I think for the Mac Pro just based on Apple providing 1.5TB of RAM support with the current iteration that an EPYC processor would be most appropriate as the current Threadripper models top out at 256GB of addressable memory support and do not support RDIMM's. However... AMD could easily do Apple a custom series of chips if Apple wanted it, they do it for other companies.

 

[OkiRun]

Pretty sure dude hasn’t had any “passion projects” for about nine years now. Unless I’ve slipped into a parallel dimension where he didn’t pass on in 2011?

Apologies to Cook
I thought about editing but what would be the point ~ Not like Jobs was gonna sue me.

 

[zephonic]

Maybe once they able to start making better ARM chips, why not making the ARM server as they did before? The server market is quite huge and that's the main financial source where Intel is getting a lot of profits. Workstation is such a niche market but not a server since people need to pay tons of money for making a server.

You mean like Xserve? There was a lot of indignation (even on this forum) when Apple discontinued them, and some intrepid soul emailed Steve Jobs demanding to know why. He replied in his customary terse way: “they weren’t buying them”

That’s just not a good fit for them. Apple’s strengths are design and UI, neither of which are relevant in a data center.

 

[high heaven]

You mean like Xserve? There was a lot of indignation (even on this forum) when Apple discontinued them, and some intrepid soul emailed Steve Jobs demanding to know why. He replied in his customary terse way: “they weren’t buying them”

That’s just not a good fit for them. Apple’s strengths are design and UI, neither of which are relevant in a data center.

Well, we are in a different situation than before. ARM is rising especially for the server. Ampere made 80 cores ARM chip and others too. But obviously, macOS itself isn't good for the server system so maybe they need improve both CPU and software?
[automerge]1592112668[/automerge]

 

[zephonic]

It's not about ARM, it's about where Apple is now as a company. They have become a luxury brand with swanky retail spaces, their products look great and are easy to use (generally).

None of that matters for data center customers. They need performance per watt at the best price possible. Apple wouldn't be able to get their normal high markups on their hardware, as their traditional strengths mean diddly-squat in that market.

They'd have to REALLY want to be in that market, and I can't see a good enough reason why they would.


But hey, I've been wrong about this stuff before, so who knows?

 

[Quu]

It's not about ARM, it's about where Apple is now as a company. They have become a luxury brand with swanky retail spaces, their products look great and are easy to use (generally).

None of that matters for data center customers. They need performance per watt at the best price possible. Apple wouldn't be able to get their normal high markups on their hardware, as their traditional strengths mean diddly-squat in that market.

They'd have to REALLY want to be in that market, and I can't see a good enough reason why they would.

Mhm exactly. I couldn't agree more with everything you're saying here.

 

[dannys1]

That would mean they'll have to maintain two separate architectures on Macs in the market at the same time which to me just seems extremely unlikely.

Not only would it increase development effort for both Apple and third party developers but it would also create inconsistencies between platforms. Already today there is a wide spectrum of x86 based Intel processors being used in Macs with features some processors have and some do not.

For example some processors have integrated graphics and offer Intel QuickSync which is a video encoder/decoder that is highly efficient and desirable. It can accelerate H.264 and H.265 encoding and decoding while using very little of the normal cores present in the Intel CPU's which saves power and reduces heat while being much faster.

This feature is not present on the iMac Pro or the Mac Pro because these processors do not have integrated graphics in the CPU package and thus do not contain Intel QuickSync. Contrastingly these Macs contain processors which feature the AVX-512 instruction and that is not present on the normal iMac, MacBook or the MacBook Pro's.

This instruction extension can accelerate performance for scientific simulations, financial analytics, artificial intelligence (AI)/deep learning, 3D modeling and analysis, image and audio/video processing, cryptography, and data compression.

These are just a couple of differences I chose because they're easy to convey but if you look at the iPhone for instance it has a dedicated area of transistors for machine learning. This area of compute is growing rapidly as it becomes used for more and more things, it's not just important for facial recognition and self-driving cars.

Eventually your computer could leverage this for better compression of files, better blocking of ads on the web, better searching of files on your computer, better noise suppression on your microphone etc

Since Apple is already pressing forward with this on their A series chips it stands to reason they would want to bring the same innovations to the Mac and that would mean a full top to bottom processor rollout to keep hardware features consistent in my opinion.

This inconsistency on the Intel processors has caused Apple consternation on multiple occasions for instance when they first released Sidecar for the iPad and some Macs didn't support it due to the need for efficient H.265 video encoding. Apple eventually included such an encoder in the T2 security chip as a way to make all Macs (iMac Pro, Mac Pro etc) consistent in this feature availability, something they wouldn't need to do if they made the entire SoC themselves.

I don't disagree with anything you've written here and in theory yes maintaining two repos would be more work. However I half expect the ARM report to use all the work from iOS cores.

I'll believe it when I see it but I just can't imagine Apple building an ARM chip that can compete with Threadripper or even Intel for top end machines - I feel it's much better suited to lower end systems where they can destroy Intel's M3 type offerings.

 

[Quu]

I don't disagree with anything you've written here and in theory yes maintaining two repos would be more work. However I half expect the ARM report to use all the work from iOS cores.

I'll believe it when I see it but I just can't imagine Apple building an ARM chip that can compete with Threadripper or even Intel for top end machines - I feel it's much better suited to lower end systems where they can destroy Intel's M3 type offerings.

It certainly won't be easy but they've shown they can deliver in the mobile arena which is actually a very hard market you may notice Intel doesn't offer any mobile phone chips anymore they simply couldn't compete, it was too hard for them. But here is Apple who doesn't even make chips swooping in with the undisputed best processor on mobile bar none.

I got confidence!

 

[PinkyMacGodess]

You mean like Xserve? There was a lot of indignation (even on this forum) when Apple discontinued them, and some intrepid soul emailed Steve Jobs demanding to know why. He replied in his customary terse way: “they weren’t buying them”

That’s just not a good fit for them. Apple’s strengths are design and UI, neither of which are relevant in a data center.

There was a problem with them. SETI had a pair of them in Puerto Rico, and the engineers there had high hopes for them. They were, on paper, faster than any other system they were using, but in process, they were slower, and they couldn't figure out why. Engineers and programmers from Apple had been flown there to see the systems in operation, and try to figure out where the problem was. At the time I was there, they still didn't have a clue, and the systems sat unused. SETI was very frustrated, as can be imagined, being saddled with something that could really help them, that was failing them. They servers likely went back to Apple. Weird...

 

[Slash-2CPU]

Apple just announced a trade-in program for Macs. That can be taken to mean they intend to support Intel Macs for a good while. It could also mean they want to eliminate all the Intel Macs asap and are willing to subsidize customers getting rid of them.

 

[Macintosh IIcx]

Apple just announced a trade-in program for Macs. That can be taken to mean they intend to support Intel Macs for a good while. It could also mean they want to eliminate all the Intel Macs asap and are willing to subsidize customers getting rid of them.

If they want developers to rewrite software for Arm macOS they need a decent critical mass of those new Macs. Otherwise developers will just postpone it as long as they can and we get nowhere. So I would think option 2.

 

[tanoanian]

If they want developers to rewrite software for Arm macOS they need a decent critical mass of those new Macs. Otherwise developers will just postpone it as long as they can and we get nowhere. So I would think option 2.

Why rewrite macOS for ARM when you have a perfectly optimized iPad OS for ARM? Apple is already maintaining two separate codebases. One for ARM, one for Intel.

 

[Macintosh IIcx]

Why rewrite macOS for ARM when you have a perfectly optimized iPad OS for ARM? Apple is already maintaining two separate codebases. One for ARM, one for Intel.

Maybe Finder for ARM as the bridge? ?

 

[StuAff]

A couple of points we can be fairly certain of...
1. Apple is fed up with Intel's troubles over the past few years, patience has not been a virtue in this situation, and it's run out.
2. If it's decided to make a switch, it won't be for financial reasons, supply/logistics, or for device performance. It'll be all of those things. I remember all too well the problems caused when the wheels came off the PowerPC wagon, both in terms of increasingly creaky upgrades, and increasingly creaky marketing. The presentation by John Rubenstein on the 'megaherz myth' (or, "we know these G4s are 'slower' than P4s, but they're not slower, honest..."). And when the change comes, they'll want to sell it in clear, unambiguous terms, just like they did with going to Intel in the first place. Faster? Lower power consumption and temperatures? Cheaper (yes, early Intel systems were cheaper than PPC, admittedly thanks to the terms of the deal, and they'll likely keep margins the same if it means a more attractive price)? Better future roadmap? Yes.

As I stated up thread, were it me I'd want Apple to pick and choose suppliers on a product by product basis. No reason why they couldn't use (for example) EPYC for the Mac Pro, i7/i9 for MBP, and ARM for the Air. Apple has in the past used multiple GPU suppliers, it probably would now if not for the nVidia fallout, nothing stops them using Intel, A-series, and AMD CPUs if they so desired. Unlikely, but they could, and it would still be an option for the future. But we'll see next Monday…

 

[pasamio]

Why rewrite macOS for ARM when you have a perfectly optimized iPad OS for ARM? Apple is already maintaining two separate codebases. One for ARM, one for Intel.

The kernel that both OS' share has support for both Intel and ARM, the makefile for XNU has referenced ARM since 10.5 (same year as iPhone) so it would be a single code base and if there is platform specific code or instructions those are compiler guarded. They likely won't rewrite a significant chunk of macOS for ARM in the sense that they've already done that. They've implemented a lot of the low level APIs in the last few years to share them between iOS and MacOS (e.g. Metal) so those APIs already have implementations in both architectures where it applies. They've been rowing in this direction for a while and if you're using a lot of these Apple provided modern frameworks then I would expect the porting experience will be relatively smooth.

The flip side is that they've been adding new API extension points to the operating system and starting to get rid of direct kernel extension pathways. They've flagged the removal of OpenGL support and they also shook up the ecosystem by removing 32-bit i386 support with Catalina. For a while now they've been introducing 64-bit only APIs which should have also been a signal to folk years ago that 32-bit was going away (keeping in mind the G5 was a 64-bit architecture as well, it's not new).

I think you'll find they're not maintaining two code bases but instead two code paths in places where low level support is different between architectures. For the majority of their code base I expect they've got it working on two platforms without duplicating code.

 

[high heaven]

We know nothing about ARM Mac and it might be different than what we are expecting. I would wait until next week to see what is ARM macOS will be.

 

[djjeff]

2. If it's decided to make a switch, it won't be for financial reasons, supply/logistics, or for device performance. It'll be all of those things. I remember all too well the problems caused when the wheels came off the PowerPC wagon, both in terms of increasingly creaky upgrades, and increasingly creaky marketing. The presentation by John Rubenstein on the 'megaherz myth' (or, "we know these G4s are 'slower' than P4s, but they're not slower, honest..."). And when the change comes, they'll want to sell it in clear, unambiguous terms, just like they did with going to Intel in the first place. Faster? Lower power consumption and temperatures? Cheaper (yes, early Intel systems were cheaper than PPC, admittedly thanks to the terms of the deal, and they'll likely keep margins the same if it means a more attractive price)? Better future roadmap? Yes.

Apple moved to x86 because PPC longevity was in question. It also had the benefit of permitting the Macintosh to natively run Windows and its huge library of software along with other x86 based operating systems. The cost to develop processors is extremely high and it became apparent production of the PPC processor was difficult to justify.

I'm not convinced a switch to ARM is going to result in a processor that is substantially faster than x64 and, if it does, what longevity such a lead would have. It'll be interesting to see.
[automerge]1592159638[/automerge]

It certainly won't be easy but they've shown they can deliver in the mobile arena which is actually a very hard market you may notice Intel doesn't offer any mobile phone chips anymore they simply couldn't compete, it was too hard for them. But here is Apple who doesn't even make chips swooping in with the undisputed best processor on mobile bar none.

Intel could make their own ARM processor.

 

[StuAff]

Apple moved to x86 because PPC longevity was in question. It also had the benefit of permitting the Macintosh to natively run Windows and its huge library of software along with other x86 based operating systems. The cost to develop processors is extremely high and it became apparent production of the PPC processor was difficult to justify.

They're still making Power architecture CPUs, currently on POWER9…And Intel's struggles to improve its designs are well-documented. AMD is making a significant dent in Intel's market dominance, because they keep making better chips at lower prices. Intel will catch up, eventually, but its current performance is resting on past laurels. Whatever Apple's going to do, it's not going to accept this for another few years.

I'm not convinced a switch to ARM is going to result in a processor that is substantially faster than x64 and, if it does, what longevity such a lead would have. It'll be interesting to see.

There's a term much used in financial ads over here: 'Past performance is not an indication of future results'. Which is true here, to an extent. Trying to use Windows on ARM as a benchmark is pointless, because Microsoft hasn't really bothered optimising it, macOS ARM certainly will be, and none of the ARM chips for the platform are on a par with Apple's. Let alone ARM chips with heftier power envelopes and desktop-grade cooling…