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

[savagemic]

Anyone that is spending $8k-40k on a computer is making up for it somehow. If it's discretionary spending, then they have it to burn anyway and are making a hell of a living.
[automerge]1591800059[/automerge]

I'm a big enthusiast, build my own water cooled computers and servers which I have done so for over 15 years. I know a great deal about computer hardware right down to manufacturing including chip fabrication and design.

I'm also a software engineer and I write server based software that runs on x86 and ARM based architectures, in the past I worked on FPGA hardware where we prototyped chip functions before committing to designing and fabricating ASIC's. I say all this to provide some context to what I'm about to say next.

Based on all the relevant data we have about Apples ARM chips they are core-for-core already better than Intel on generalised computing tasks. And that in itself is absoloutely insane because the chips Apple is producing are in phones with extremely small power and thermal budgets.

We're talking execution parity with processors that consume 30 to 50x more watts and release 100x more heat. This is literally unheard of in the industry and to put this in perspective for Intel to even get close (and still be 10x worse) they would have to bin their processors to within an inch of their life.

If you're not familiar with that term it's where they test the chips they produce and put them in different categories based on their performance such as how high they clock, if any parts of the chip are defective, how much power the chip consumes at the clock speeds it's capable of, how much heat it produces while under load etc

Apple has managed to reach parity not with their top 0.01% of produced chips like Intel does with some of their super high-end XEON's and ultra low-power U skus. They've been able to do it with from what we can decipher 86-92% of their entire chip yields.

They sell 100 Million iPhones with their latest SoC every year and they all perform the same which is to say absoloutely steller, top of the line performance, two years ahead of their closest mobile counterpart (Qualcomm Snapdragon, Samsung Exynos etc).

So with all that out of the way what does all this mean for the Mac Pro? - When you're wanting to build a large chip one with a wide memory bus, lots of cores, lots of on-board cache you need to start with a strong foundation and in todays chip fabrication that means more than anything performance-per-watt.

If you have a die that consumes 2.5 Watts per core and you scale that to 64 Cores which is the kind of chip appropriate for a next generation Mac Pro suddenly you have 160 Watts for raw core compute. And that's before you factor in the power consumption for core-to-core communication (uncore) which with that many cores could be 20-30 watts then the I/O such as memory and PCIe and any other "uncore" usage.

Things can quickly spiral into the 250-350 Watt range. But here is Apple with an architecture that is already sub 1 Watt for the cores. Suddenly they can produce a 64 Core chip where all the cores can be 60 Watts leaving ample room for uncore power.

This is what's exciting. Instead of coming at the processor design challange from the top (performance) they've come at it from the bottom (low power). This lends itself perfectly to making a large chip with lots of cores a chip that is appropriate for a Mac Pro class computer.

Now on top of this as I mentioned Apple has put these high performance chips in phones that have very small thermal envelopes and yet we've seen Apple able to reach very high clock speeds on these processors even when under sustained loads. This is noteworthy because this is without proper heatsinks. Apple at most has an IHS on their chips now (Integrated Heat Spreader) which is thinner than the thickness of a coin.

When they design a chip using the principle architecture of their mobile SoC's into laptops and desktops where they can attach heatsinks that have 45 Watts (MacBook Pro) to 300 Watts (Mac Pro) of heat dissipation they can run them a lot faster.

Based on AMD's usage of TSMC's 7nm process node we know that the high performance node offered by TSMC (which does differ slightly from the low-power 7nm node utilised by Apple) that the sweet spot for the transistor switching speed is around 4GHz.

This is the point where heat output, power consumption and clock speed come together for the best trade offs on each to deliver a high performance chip which doesn't guzzle energy essentially. So right now in a phone Apple is pushing 2.3GHz and already streamrolling Intels 3.6GHz-4.2GHz mobile chips core-for-core. Now imagine what Apple can do delivering their own archicture at these same clock speeds.

I need to remind you, we can only do projections because we can't overclock an iPhone SoC to see what might be.

Now I do want to temper expectations a little. There are things Apple has to overcome to deliver for a Mac Pro type computer.

Firstly I cannot overstate how difficult it is to keep so many cores fed. The interconnects between CPU cores in a single die can really hamper performance especially in the kinds of high end workflows professionals will be performing where core-to-core communication is highly utilised due to multithreading.

Secondly if Apple decides to make chips that are one huge die (like Intel) that will decrease yield rates due to increases in defects. It will also increase costs as more of the wafers produced for them will go to waste. So this is a two-fold problem, clock speeds and core counts may become restricted with this strategy.

They could potentially go with a multi-die setup similar to AMD's Zen1 or Zen2 where by you make smaller dies that are all identical and combine them together on a single module to create the CPU. If Apple were to do this it would allow for higher yields, higher clock speeds (especially on the high core count part appropriate for a Mac Pro) and lower their costs.

Thirdly scaling up an entire chip for a desktop takes time. There is a lot of engineering they can't just skip over, there is stuff they haven't done even for the iPhone and iPad. For example their current SoC has PCIe lanes and they use NVMe storage on the iPhone and iPad. That's great when you only need 4 lanes but the Mac Pro for example needs 72. This means Apple has to decide do we put 72 or more PCIe lanes into our SoC or do we put say 32 and use PCIe switching chips? - There's trade offs. Also do they move to PCIe 4.0 or even 5.0 (2021 5.0 will be making the rounds in shipping systems from their rivals).

Forth and perhaps the most important. While Apple is dominating in Mobile and their performance eclipses Intel currently (when normalising core count and frequency) there is another chip manufacturer on an unbelivable climb to the top and that is AMD.

What happens if AMD is faster than Apple and they made everyone do all this work switching? What if Apple can only produce a 32 Core part for their first Mac Pro refresh when AMD already today is selling a 64 Core chip? What if Apple can only deliver 72 PCIe lanes or the lanes they do produce are only PCIe 3.0 when AMD today is delivering 128 PCIe 4.0 lanes.

That is to me the biggest gambit they're taking here. In 2021 when these new ARM macs are supposedly coming out AMD will be delivering Zen4 based EPYC processors that feature 128 or more PCIe 5.0 lanes and 64 Cores+ (the rumour being 72 Cores but if they move to 5nm by then it may be as much as 128 Cores, all over 3GHz and under 250 Watts power consumption).

Anyway, interesting times. Personally if Apple can deliver something that is better then I say they should go for it. I don't think the downsides of software compatability headaches should hold back chip progress.

I'm not a chip or software expert but this was a fantastic read. Thanks for taking the time to share.

 

[deconstruct60]

....

Apple has proven they can design some really kickass ARM based CPUs, they are far ahead of the competition in the mobile market. If they can take that expertise and apply it to bigger designs that can be actively cooled, I'm excited to see what they can do!

What Apple hasn't shown much of over the last 4-6 years is that they can walk and chew gum at the same time. Big updates to Mac product X means that products Y and Z largely lie fallow (or maybe minor tweaks). The A-nnX processors are 18-24 month leaps waiting on process shrink to move forward. A12Z is just lighting up a GPU core that was present but just not turned on in the initial roll out of A12X.

If they shifted most of the resources over to doing something else... sure could do something else. Like to $2,500-3,500 Mac Pro ... Apple could have done that, but they didn't.

The Bloomberg article points out what Apple's initial primary interest is here as the representative example of why putting in this effort. Thinner and lighter laptops. Is that laser focus really going to get you a better Mac Pro than would could 2021-22 Workstation focused CPU package from Intel or AMD ?

 

[quagmire]

Just curious... What were you projecting on how long would it take for a ROI? As stated, even without an ARM transition Apple will typically EOL a machine 5 years after they stop selling it. The only unknown is how long the 7,1 will go on without being replaced by an updated model. That can range from 1-2 years to 6 years( if we go by the trash can longevity).

 

[jasonmvp]

Is that laser focus really going to get you a better Mac Pro than would could 2021-22 Workstation focused CPU package from Intel or AMD ?

Why yes, yes it will! That is, if you're to follow along with the ARM proponents, of course.

(I still think this is a massive mistake on Apple's part)

 

[Rudy69]

What Apple hasn't shown much of over the last 4-6 years is that they can walk and chew gum at the same time. Big updates to Mac product X means that products Y and Z largely lie fallow (or maybe minor tweaks). The A-nnX processors are 18-24 month leaps waiting on process shrink to move forward.

The people they have working on the CPUs wouldn't be working on anything else regardless. They employ ~137,000 people, I think they'll manage and if they need to they will hire the required experts (here's an example: https://www.theverge.com/2019/6/26/18760083/apple-arm-architect-hire-cortex-a76-mac-processors-intel)

 

[jasonmvp]

The people they have working on the CPUs wouldn't be working on anything else regardless. They employ ~137,000 people, I think they'll manage and if they need to they will hire the required experts

I sort of think you missed deconstruct's point. It's not just the chip or hardware specialists that are needed. The company as a whole has to decide to walk and chew gum at the same time, to use his phrase. And it really seems they're not capable of doing it. Or they don't want to. Just having the hardware specialists in house doesn't mean the product managers will be given the go ahead to push forward. Or push forward with any real speed.

 

[AndenIDK]

ARM would mean I would ditch Mac computers in general

 

[EdT]

I think I understood about 51% of what Quu wrote, but I think it highlights that Apple Arm chips are very powerful, not just for phones but compared to Intel chips in computers, that the problems users had when Apple switched from the PowerPC processor aren’t as likely because Assembly or machine code to get improved software performance isn’t as necessary today as it was in the middle 2000’s. I would THINK but I don’t know that drivers for components (screens, drives, ports) are still written in machine or assemblers but I haven’t messed with adding components, I just plug whatever into a USB port so maybe that isn’t an issue.

 

[th0masp]

Actually it is rude, and quite presumptuous. Thanks.

One way to look at it is that if that transition is going to happen then for a good few years you will be holding on to the only upgradeable piece of hardware built for running the proven platform - the one where you know the software indeed works. These transitions tend to be .... interesting for the end user of the latest & greatest gear.

But definitely you will be left behind at some point with deprecated software. No chance of getting the mileage out of this as you would have from the previous MPs.
Assuming that this transition is indeed happening of course.

I'd return it! It's an obscene amount for a personal computer if that's what it is to you. Surely if you want to stay with Mac you can find an interim replacement from among their current products that is not going to damage the wallet quite this badly and then lean back and watch the transition take place. Pick up a new style ARM-Mac way down the road, probably save some money overall.

 

[bsamcash]

Would ARM be able to support Thunderbolt?

 

[Azrael9]

I would imagine apple's new processor would eventually apply to their whole mac lineup. Wondering if that means Mac Pro 7,1 will be phased out sooner or later? Perhaps my concerns don't make sense, but I wanted to ask the board their thoughts around the future of the Mac Pro with apple's own ARM processor.
[automerge]1591753616[/automerge]

When Apple decides to bury the Mac Pro with Intel?

It will. Bury INtel standing up.

I wouldn't be investing 40k into a Mac Pro right now.

Unless I've got money to burn and can earn that money easily with 'said' machine.

I think any A16X chip in two years time finishes the Mac ARM transition. ANd finishes off the Mac Pro intel.

Azrael.

 

[dspdoc]

Get over yourself. Anyone that is spending $8k-40k on a computer is making up for it somehow. If it's discretionary spending, then they have it to burn anyway and are making a hell of a living.
[automerge]1591800059[/automerge]


I'm not a chip or software expert but this was a fantastic read. Thanks for taking the time to share.

No, you get over yourself! Don't presume to know people's finances and how said finances may or may not be recouped. You're completely missing the point anyway that this kind of investment should have some longevity regardless of the ROI. Have a nice day! ?‍♂️
[automerge]1591804036[/automerge]

When Apple decides to bury the Mac Pro with Intel?

It will. Bury INtel standing up.

I wouldn't be investing 40k into a Mac Pro right now.

Unless I've got money to burn and can earn that money easily with 'said' machine.

I think any A16X chip in two years time finishes the Mac ARM transition. ANd finishes off the Mac Pro intel.

Azrael.

That's quite a lofty assumption. LOL
[automerge]1591804229[/automerge]

One way to look at it is that if that transition is going to happen then for a good few years you will be holding on to the only upgradeable piece of hardware built for running the proven platform - the one where you know the software indeed works. These transitions tend to be .... interesting for the end user of the latest & greatest gear.

But definitely you will be left behind at some point with deprecated software. No chance of getting the mileage out of this as you would have from the previous MPs.
Assuming that this transition is indeed happening of course.

I'd return it! It's an obscene amount for a personal computer if that's what it is to you. Surely if you want to stay with Mac you can find an interim replacement from among their current products that is not going to damage the wallet quite this badly and then lean back and watch the transition take place. Pick up a new style ARM-Mac way down the road, probably save some money overall.

Yes, I am strongly considering the mini with an eGPU. I will certainly miss having all the components such as hard drives inside the machine however. With a mini it tends to be dongle city and lots of external devices plugged in. With the 7,1 everything is nicely tucked away inside the machine.
[automerge]1591804255[/automerge]

Would ARM be able to support Thunderbolt?

EXCELLENT Q! I have loads of TB3 devices.
[automerge]1591804354[/automerge]

ARM would mean I would ditch Mac computers in general

I am genuinely curious as to why...?

 

[deconstruct60]

It means you need to start thinking about either moving your hack to Windows or purchase a real Macintosh. The Hackintosh movement has about 2-4 years left before it disappears. Ultimately, it depends on how long Apple supports macOS on Intel, but it won’t be more than 5 years and probably less.

Two pretty big assumptions buried there. One, that Apple transition the whole entire Mac line up quickly. Two, that they largely ignore 'fat binaries' and app store application distribution.

The first is pretty weak given what it has been in the Bloomberg leaks about this. It very much sounds like Apple has one single processor. And that over a couple to several years they will roll out more processors. Nothing here sounds like they are going to get up on stage and promise they will be done in 12 months. ( 14 months or 18 months).

As long as the transition takes more than a few years that "5 or less" countdown clock probably won't start. Apple has to stop selling new x86 based Macs to pragmatically start a count down clock.

The article talks about processors over time. Derived on A14 baseline in 2021 . Derived on "next generation" ("a-15?") lined up. etc.

"...Apple has already started designing a second generation of Mac processors that follows the architecture of chips planned for the 2021 iPhone. That indicates Apple wants to put its Macs, iPhones and iPads on the same processor development cycle. ... "

Apple Aims to Sell Macs With Its Own Chips Starting in 2021

Apple Inc. is planning to start selling Mac computers with its own main processors by next year, relying on designs that helped popularize the iPhone and iPad, according to people familiar with the matter.

www.bloomberg.com

there is a big gap between doing a "Mac" processor and doing a whole broad line up of processors for Macs. Tightly coupling it to the iPhone processor doesn't really point to that. And frankly the iPad Pro processor is detached slightly from the iPhone processors. The notion that a even broader "Mac" processor line up wouldn't get even more decoupled than that is suspect ( largely for same reasons. huge volume differences and process shrink iteration times ).


Second, that the vast bulk 100M x86 Macs are suddenly going to be decommissioned and disappear in 5 years. Larger PC upgrade cycles have been getting longer now. For Macs also ( Windows based ones have been longer but Macs are not like it was back in 2006-8 era a decade or so ago. ). Dropping some relatively super low volume, big expensive Mac Pro isn't an issue driving away a significant chunk of the user base.

Depends upon what Apple does here with the ARM macs. They could use them to substantively grow the Mac ecosystem. Sub $900 laptops and small desktops that are useful in expanding Apple's footprint in more developing economies where Macs are just plainly not competitively priced at all. Or Apple can shift to holding the same sub 8% worldwide market share in classic general (laptop/desktop) PC form factor space and just keep the same pricing they have now. If looking to make the most money on 5% market share of much higher margined system then uniform top-to-boom own ARM may make sense. (more money from fewer folks). The latter though really do much to encourage iOS folks to port to a small ( and perhaps shrinking) market even if lower porting costs incrementally with a "easy to port" API.

If Apple tries to actively herd folks onto newer, more expensive equipment, then probably will get some folks to bolt. ( certainly have for the Mac Pro move they have made to 100% higher entry configuration costs. )

If WWDC has themes of "getting your apps to ARM with an App store deployments is sooooooo much easier and better" then the transition time probably won't be as short as folks think it will be. ( getting a larger piece of the services action will make Apple happy. )

 

[Spanky Deluxe]

If they do go ahead with full ARM across the board then I have to believe that Apple would only do that if they think their computers can be *faster* than what Intel offers and they have to believe that they'll be faster for years to come. They'll probably create something like Rosetta to handle the handover, which went incredibly well. The only real downside will be Windows compatibility and virtual machines. I imagine Apple has a good idea of how many people use BootCamp. It's probably a pathetically small proportion of users and not worth worrying about. Virtual Machines is a bit more tricky though because they're frequently used by developers and other more 'high end' users.

For people with lower demands, they'll likely be able to use emulation software and the performance will be 'fine'. People are buying Apple's MacBooks with processors that are slow compared to 5 year old machines, a slow virtual machine probably won't matter to a lot of people. For higher end users, performance can be a real problem. However, there are already solutions for having servers that just run virtual machines that are likely more cost effective than a Mac Pro. Alternatively, I can imagine a market for x86 daughter cards specifically for virtual machines springing up, potentially provided directly by Apple or by third market providers. I imagine a lot of Mac Pro buyers who need the occasional virtual machine would be just fine with spending an extra $1000 on an x86 daughter board that can run virtual machines without sapping any power from the main machine.

 

[laylow]

That's still an EXTREMELY short lifespan for those that dropped north of $40k! I have $8k into mine, which IMHO is still a boatload for a computer. I guess it's all relative and depends on perspective. For the stupidly wealthy, that is nothing. For the everyday working man, that's quite an investment!

Yep...but let's be honest, Apple bones us like that sometimes.

 

[deconstruct60]

Would ARM be able to support Thunderbolt?

The question is more so is Apple going to put in the work to support Thunderbolt firmware and boot support. ( or hire someone to do it). Thunderbolt never was dependent upon some specific instruction set. It always worked in peripherals that have no x86 CPU in them at all.

If could have worked with AMD x86 processors too if AMD had put in the work. They didn't want to.

Is Apple going to do their own Thunderbolt implementation. I suspect not. At least not for a couple of years. Most likely there will be some 3rd party discrete chipset implementation eventually. On the initial stab at a Mac ... perhaps not if content to just do a revival of the one port wonder Macbook.

 

[dspdoc]

Yep...but let's be honest, Apple bones us like that sometimes.

Indeed they do! Often...

 

[Detektiv-Pinky]

Better performance is great, however there are a lot of other things to consider at well.

Many people use MacPro to get access to Windows via VMs or bootcamp. This makes Mac a flexible development platform. A move to ARM could change this.

Important software is another factor. Some major players, like Adobe, do not have excellent track record to do quick and efficient platform changes. For a lot of niche programs the future platform change be a bit uncertain.

You are exactly describing, why I find this move troublesome.

I guess a lot of independent software developer are more or less limping along and the macOS app-store has not been the overwhelming success story that Apple likes it to be. It took several years for certain specialized applications to make the move to the Mac. Not everything is a Photoshop or MS Office. So, there is a real danger that smaller developers and specialised software that require their own frameworks will be thrown under the bus by this move...

And all for what? A few points in a synthetic software benchmark, that may be almost meaningless in real-world applications?

 

[fastlanephil]

It means you need to start thinking about either moving your hack to Windows or purchase a real Macintosh. The Hackintosh movement has about 2-4 years left before it disappears. Ultimately, it depends on how long Apple supports macOS on Intel, but it won’t be more than 5 years and probably less.

I imagine Macstorm isn’t too thrilled about this news either. Apple probably won’t need to incur attorney fees like when they did running Psystar out of business.

Macstorm is a Krakow Poland based hackintosh builder currently with multiple product listings on eBay that ships to the US.

 

[deconstruct60]

If they do go ahead with full ARM across the board then I have to believe that Apple would only do that if they think their computers can be *faster* than what Intel offers and they have to believe that they'll be faster for years to come.

It isn't about faster.

"... Apple’s chip-development group, led by Johny Srouji, decided to make the switch after Intel’s annual chip performance gains slowed. Apple engineers worried that sticking to Intel’s road map would delay or derail some future Macs, according to people familiar with the effort. ..."

Apple Plans to Announce Move to Its Own Mac Chips at WWDC

Apple Inc. is preparing to announce a shift to its own main processors in Mac computers, replacing chips from Intel Corp., as early as this month at its annual developer conference, according to people familiar with the plans.

www.bloomberg.com

There is a bunch of 2015-2017 thinking there that Intel is the only viable x86-64 option.... otherwise they'd have to whole thing themselves. In 2020 that is a bit more suspect with current information of solution space available.


And also about doing something with the chips to reach some Apple external container design goal .

"... . Apple’s processors are also more power-efficient than Intel’s, which may mean thinner and lighter Mac laptops in the future. ..."

Ultimate Power probably isn't the driver here. Part of the "faster than INtel" crowing they are doing here is

"... Inside Apple, tests of new Macs with the Arm-based chips have shown sizable improvements over Intel-powered versions, specifically in graphics performance and apps using artificial intelligence, the people said. ..."

So faster than Intel iGPU. When even Intel has faster than 4 year old baseline design graphics coming.


And faster than Tensor unit inside of Intel chip. Of which there is none so ... well yeah . And again likely not really compared against the 2020 era bfloat and computation.

Intel Gen12 Xe Graphics Leads AMD's Vega in Integrated GPU Performance

The tide has turned and roles have swapped, with AMD now delivering more CPU but less GPU.

www.tomshardware.com

Dollars to doughnuts when Apple gets on stage and starts chest beating it will be about dated processors. ( like comparing Mac Pro 2019 to Mac Pro 2013. or the A12X GPU to Xbox One X GPU performance. ).

They'll probably create something like Rosetta to handle the handover, which went incredibly well.

They didn't create something like Rosetta when the rolled out Rosetta. Which x86-64 AVX-512 capable emulator-JIT is suppose to fall out of the sky to save them this time should prove interesting. Especially if they are kicking up a dust storm of Intel 'Hate' dust up as stomp away.

 

[bsamcash]

The question is more so is Apple going to put in the work to support Thunderbolt firmware and boot support. ( or hire someone to do it). Thunderbolt never was dependent upon some specific instruction set. It always worked in peripherals that have no x86 CPU in them at all.

If could have worked with AMD x86 processors too if AMD had put in the work. They didn't want to.

Is Apple going to do their own Thunderbolt implementation. I suspect not. At least not for a couple of years. Most likely there will be some 3rd party discrete chipset implementation eventually. On the initial stab at a Mac ... perhaps not if content to just do a revival of the one port wonder Macbook.

Don't both Apple and Intel own the licensing rights?

 

[Azrael9]

That's still an EXTREMELY short lifespan for those that dropped north of $40k! I have $8k into mine, which IMHO is still a boatload for a computer. I guess it's all relative and depends on perspective. For the stupidly wealthy, that is nothing. For the everyday working man, that's quite an investment!

Over 5 years. 8 grand a year for a 40k Mac Pro. Ouch.

Be better if they offered rational, affordable Mac towers.

Azrael.

 

[anthony13]

I work in an industry that at times relies on some pretty niche applications and software from a multitude of developers. I remember the pains of the software switch in 2005-2008 and am not looking forward to that at all.

 

[ApfelKuchen]

What if Apple looks at the 7,1 as a stopgap meant to hold onto power users until they can roll out a long-planned ARM-based powerbox?

Not saying there's evidence one way or the other for that scenario, just trying to stretch imaginations a bit. Folks are so focused on "What is Apple going to take away from us this time" that it's worth wondering whether Apple might not mind grabbing for a bigger piece of the low-volume/high-margin end of computing, perhaps moving beyond media production into scientific/engineering as well. A price/performance powerhouse that might tempt some major developers in those areas to make the leap.

 

[Zdigital2015]

Would ARM be able to support Thunderbolt?

Thunderbolt is a support chip on the motherboard connected via PICe to either the CPU directly or indirectly via the PCH. There’s nothing prevent Apple from implementing Thunderbolt. Whether USB4 and Thunderbolt 4 come to the Mac is dependent upon how Apple implements PCIe, which is likely to be PCIe 4.0 at this point. The T2 chip functions will undoubtedly expand is it becomes the de facto PCH for the Arm Mac.