Apple's Path to Arm-Based Macs Could Start With a New 12-Inch MacBook

[B-Mc-C]

Yes, it may outperform on some benchmarks but people who need mac to do mission critical work might want to keep their intel macs while people beta test the ARM macs. It makes sense to start with 12inch Macbooks and not the Pros.

My point exactly.

 

[matram]

I expect 20% improvement > than each intel part an apple arm is replacing.

Based on your expertise could you elaborate on how you believe Apple would achieve this.

Increased clockspeed, relying on better thermal efficiency?

Changes to the CPU architecture to optimize for high performance?

Or something else?

 

[cmaier]

Based on your expertise could you elaborate on how you believe Apple would achieve this.

Increased clockspeed, relying on better thermal efficiency?

Changes to the CPU architecture to optimize for high performance?

Or something else?

1) much better process node. Intel 14+++ just ain’t cutting it. And their 10 isn’t in much yet. Apple’s chips will be at least one process node generation ahead.

2) better designers (for a long time at AMD I was in charge of design methodology. i evaluated lots of different ways to design chips - we had EDA vendors come in and I’d hand them a part of a logic block and ask them to take their best crack at designing it. Without fail, every single time, they’d come back about a month later and hand me a design that was either 20% slower than what our designers did, or 20% more power-hungry, or 20% more area. Or some combination of those. The design methodology we used was distilled from the best practices at DEC (the Alpha team), and Exponential (the PowerPC x704 team), and NexGen (the original K6 team), who all had very similar thoughts on how chips should be designed. Remember that Intrinsity (which ended up part of Apple) was an Exponential spin-off, and PA Semi was a bunch of ex-DEC folks, and many of the people I worked with at AMD are now at Apple. These folks know what they are doing.

In the days of Opteron I attended a conference in Monterey. A bunch of Intel people were there. I heard them talking about our chip, which had been announced but not yet on sale. They were talking about how perplexed they were that we could possibly have done such a thing.

Do not underestimate the difference good logic, circuit and physical designers can make.

3) inherent advantage. No matter what, all else being equal, RISC has an advantage. The amount may vary, but x86 adds about 20% to the size of a typical core (could be less - depends on how many pipes/ALUs, etc.). You always have to have more stages to deal with things that instructions *might* want to do, but rarely do in practice, just to maintain compatibility, etc. If this was the *only* issue, the difference in performance could be pretty small. But in the hands of good designers, on a good fab process, this difference is probably magnified.
[automerge]1592201231[/automerge]

Gruber's Daring Fireball blog is quoting Brendon Shanks hypothesis that it's going go be the A12Z SOC from the iPad Pro inside a Mac Mini or Apple TV chassis, which requires less engineering to create a developer-only box than a full MacBook. If so, I wonder if the A12Z can run at a higher thermal envelope then when wedged inside an iPad? Or if it's going to be a chip in a different package with something like a bigger heat spreader? Maybe real (non-dev) ARM Mac customers will have to wait till next year for some higher wattage variants of the A13 or A14.

Dalrymple's Loop blog might be passing along a whisper that Apple already has these systems working, ready to go, and that they are "fast".

The hints are out that Apple wants developers to really pay attention to the coming Keynote and Platform sessions.
[automerge]1592196957[/automerge]


But the fastest (known) CPU is on the Voyager 2 spacecraft.

What’s weird about Shanks’ hypothesis is he thinks apple is worried about battery life getting out.

I think we’re all much more concerned about speed than battery life.

I don’t disagree that it may be some sort of brick, but I think a development laptop is at least possible (more like some sort of modified ipad, with magic keyboard case support)

 

[iZac]

The old 12" Macbook always really clashed with the Macbook Air and even more so after the new one was released.

What if the new ARM mac tried to act as a bridge device, and designed it with a floating hinged keyboard like the iPad Pro, but slimmed down. The chipset and battery can go under the keyboard rather than behind the screen, helping with the balance and allowing an F-Key row and bigger trackpad because of the better balance?

There's no logical reason to do this, other than a symbolic connection between iPad and Mac but it might be a fun way to herald in a new design direction to go with the new era.

Also - the ARM iMac could have the same hinge design, but just a 'foot' version like the current iMac.

Just a thought.

 

[cmaier]

The old 12" Macbook always really clashed with the Macbook Air and even more so after the new one was released.

What if the new ARM mac tried to act as a bridge device, and designed it with a floating hinged keyboard like the iPad Pro, but slimmed down. The chipset and battery can go under the keyboard rather than behind the screen, helping with the balance and allowing an F-Key row and bigger trackpad because of the better balance?

There's no logical reason to do this, other than a symbolic connection between iPad and Mac but it might be a fun way to herald in a new design direction to go with the new era.

Also - the ARM iMac could have the same hinge design, but just a 'foot' version like the current iMac.

Just a thought.

Since there is really no advantage to the floating design, i don’t see them doing this.

 

[matram]

1) much better process node. Intel 14+++ just ain’t cutting it. And their 10 isn’t in much yet. Apple’s chips will be at least one process node generation ahead.

2) better designers (for a long time at AMD I was in charge of design methodology. i evaluated lots of different ways to design chips - we had EDA vendors come in and I’d hand them a part of a logic block and ask them to take their best crack at designing it. Without fail, every single time, they’d come back about a month later and hand me a design that was either 20% slower than what our designers did, or 20% more power-hungry, or 20% more area. Or some combination of those. The design methodology we used was distilled from the best practices at DEC (the Alpha team), and Exponential (the PowerPC x704 team), and NexGen (the original K6 team), who all had very similar thoughts on how chips should be designed. Remember that Intrinsity (which ended up part of Apple) was an Exponential spin-off, and PA Semi was a bunch of ex-DEC folks, and many of the people I worked with at AMD are now at Apple. These folks know what they are doing.

In the days of Opteron I attended a conference in Monterey. A bunch of Intel people were there. I heard them talking about our chip, which had been announced but not yet on sale. They were talking about how perplexed they were that we could possibly have done such a thing.

Do not underestimate the difference good logic, circuit and physical designers can make.

3) inherent advantage. No matter what, all else being equal, RISC has an advantage. The amount may vary, but x86 adds about 20% to the size of a typical core (could be less - depends on how many pipes/ALUs, etc.). You always have to have more stages to deal with things that instructions *might* want to do, but rarely do in practice, just to maintain compatibility, etc. If this was the *only* issue, the difference in performance could be pretty small. But in the hands of good designers, on a good fab process, this difference is probably magnified.

@cmaier - thank you for sharing.

Having played around with Assembler on various MCU architectures back in the day I can appreciate you comment about the RISC vs CISC advantage.

I have heard several times that Intel has (had?) a good manufacturing process but bad chip designers. Now the process improvement has stalled for 5 - 6 years and TSMC has the edge. Any root cause for Intel loosing the process edge and can it be recovered or will TSMC now be the leader for the future.

 

[ds2000]

My first generation tesla model s disagrees with you.

My 2020 Model 3 disagrees with you

 

[Hieveryone]

12" MacBook is the greatest Mac of all time IMO. I've used many over the years. 13", 15", 16", and 17" and by far my favorite is the 12"

It's super light, super thin, super portable, incredibly beautiful, fully functional OS, and nothing really compares.

Everything else just feels like a laptop. The 12" MacBook is a whole new experience.

 

[aleni]

Never buy a gen 1 anything. Car, Computer.. ETC.

let the enthusiasts buy the first gen products. If all people think like you, there’ll never be any gen 2 Of any products in the world.
[automerge]1592208351[/automerge]

12" MacBook is the greatest Mac of all time IMO. I've used many over the years. 13", 15", 16", and 17" and by far my favorite is the 12"

It's super light, super thin, super portable, incredibly beautiful, fully functional OS, and nothing really compares.

Everything else just feels like a laptop. The 12" MacBook is a whole new experience.

Agreed. It’s the most comfortable laptop to use everywhere.

 

[chucker23n1]

I appreciate your well written replies, and you seem like a decent guy, but I still fail to understand all the handwringing you and others seem to have over Apple’s Mac lineup. Especially since all we are talking about here is introducing a single ultralight to the lineup that can serve as a light use laptop for people with basic needs and can easily be powered by the first ARM based Mac chip Apple rolls out.

Thing is, the MacBook Air is already the ultralight.

There's certainly bigger problems facing Apple, but yes, I think paralysis of choice is absolutely one when a company offers too many similar products.

What's your elevator pitch on whether I should get an Intel-based ultralight or an ARM-based ultralight?

Also, I’ve been recommending Macs to people since 1984, and nothing about the segmentation of their lineup has made it any harder over the past 5-10 years. The MacBook was a weird, underpowered computer, and so I didn’t recommend it to anyone. Boom. How hard was that?

Well, there were other issues, like keyboard reliability. At the end of the day, I feel like the easiest recommendation was the Air: yeah, it has an old CPU and a non-Retina screen, but its keyboard is great, it has many ports, and it's hugely popular.

Which was kind of a shame. They fixed that. Let's not get back to those days.

The keyboard thing has certainly been an issue in regards to overall quality, but that has nothing to do with their overall lineup. At one point, all the keyboards sucked.

It adds to the difficulty of making recommendations. Do you recommend the Air because it still has a more reliable keyboard, or do you recommend one of the newer MacBooks?

And, in a different way, that too is a line-up problem. Give us clear differentiation. If they put the butterfly in this new one, that almost makes sense to me. Make it way thinner and lighter even than the Air, and make it clear that its keyboard isn't the kind you get if you type a lot; it's the kind you get if you want maximum portability.
[automerge]1592208820[/automerge]

while I agree that for a small time in the late 90s the Mac lineup was simple, it quickly got complicated.
In January 1999 he introduce the power Macintosh G3 blue and white, then barely 6 months later the power mackintosh G4.
Then there was the power mackintosh G4 cube.
Then the 15 inch titanium PowerBook G4
Then the eMac G3 and the iMac G4.
Then they introduced the 12 and 17 inch PowerBook G4, but the 15 inch was still titanium while the other two were aluminum.
Then came the iMac G5, and the Mac energy forThen came the iMac G5, and the Mac mini G4.
So by 2005, there was a iBook G4, a PowerBook G4, a Mac mini G4, an eMac G4, an iMac G5, and a power mackintosh G5, with a server machine as well.
Then for a while, the lineup was pretty simple with the MacBook, MacBook Pro, Mac mini, iMac, and Mac Pro.

You're right that the 2x2 grid didn't last long, but I really don't think that was complicated. You got the Cube if you wanted a more stylish, less high-end variant of the Power Mac. You got the eMac if you wanted a cheap variant of the iMac (really, the eMac was the iMac G3 with G4-based internals). And so on.

I think it's much easier to decide between a Mac mini, eMac, iMac, and Power Mac than it was between a MacBook, MacBook Air, and 2-port MacBook Pro. All were around $1000-1500, all were laptops, all were thin and light, all had minor pros and cons. They occupied a very similar space.

 

[Hieveryone]

let the enthusiasts buy the first gen products. If all people think like you, there’ll never be any gen 2 Of any products in the world.
[automerge]1592208351[/automerge]


Agreed. It’s the most comfortable laptop to use everywhere.

Right. I even hold it in my hand and do quick tasks if need be like send off a quick email. Light enough for me to do so. Truly amazing device.

 

[playtech1]

There are many reasons which support Apple starting with a 12 inch MB replacement for their ARM Mac journey.

My view on the biggest reasons:

- lack of speed and battery life are big issues on ultraportables, which should be much better with 5nm / ARM

- Apple will be able to make comparisons with comparably sized Intel laptops, which usually run lower power processors, flattering any comparison

- on day 1 I bet not much beyond Apple's apps and some core 3rd party apps will run on this thing (at least at native speed), which is less bad for something that is marketed as super portable rather than a daily driver for professionals

- somewhat lower cost, so people might be willing to take a chance

- fills a small hole in Apple's lineup, which is also small enough that if it flops, it can be repurposed / sidelined / quietly killed without too much trouble.

I am course prepared to be shocked and find out they do a swift top-to-bottom ARM refresh!

My big worry is that I think Apple made this move because they could see Intel's stagnation, but while Apple has been preparing for ARM, AMD has forged ahead. Apple's ARM performance might not blow the x86 competition out of the water as it had hoped, which then calls into question why go through the hassle of an architecture change. Just to help Apple reduce its BoM?

 

[Gerdi]

My big worry is that I think Apple made this move because they could see Intel's stagnation, but while Apple has been preparing for ARM, AMD has forged ahead. Apple's ARM performance might not blow the x86 competition out of the water as it had hoped, which then calls into question why go through the hassle of an architecture change. Just to help Apple reduce its BoM?

AMD is not really ahead of Intel on the architecture side. They mainly have the node advantage which helps them to put more cores into their chips as well as having better power efficiency. This is what makes their chips currently a better option. But the inherent x86 issue is not solved by AMD at all. Any contemporary high performance ARM design will blow the x86 competition out of the water, be it AMD or Intel.

Take the Surface Pro X as example. It is the highest performant 7W TDP Windows device. Even the recently released Intel Lakefield SoC, which is supposed to compete in the 7W space does not hold a candle against the SQ1 of the Surface Pro X performance wise. AMD does not have any 7W option at all.

 

[ctrlzone]

while AMD has caught up, they still are inferior in key aspects, they forced Intel to lower Prices but Intel can still charge a little more. this will change with the 4000 desktop Zen 3 series at the end of this Year.

why go through the hassle of an architecture change

thats a key point, it has to blow the competition out of the water, otherwise a transition makes no sense, too much risk & hassle involved. if they cannot blow the competition, they certainly could make a low power ultra-thin device powered by ARM, not likely imo, at that point why bother with ARM.

 

[Jury]

AMD is not really ahead of Intel on the architecture side. They mainly have the node advantage which helps them to put more cores into their chips as well as having better power efficiency. This is what makes their chips currently a better option. But the inherent x86 issue is not solved by AMD at all. Any contemporary high performance ARM design will blow the x86 competition out of the water, be it AMD or Intel.

Take the Surface Pro X as example. It is the highest performant 7W TDP Windows device. Even the recently released Intel Lakefield SoC, which is supposed to compete in the 7W space does not hold a candle against the SQ1 of the Surface Pro X performance wise. AMD does not have any 7W option at all.

The same can be said for Apple SoC node advantage. However, Apple doesn't really need to do a whole architecture transition by leaving out other option that possesses great performance like AMD Zen 2 x86 for a non-mobile device like iMac.

 

[Gerdi]

The same can be said for Apple SoC node advantage. However, Apple doesn't really need to do a whole architecture transition by leaving out other option that possesses great performance like AMD Zen 2 x86 for a non-mobile device like iMac.

Not sure that you understood what i wrote. I did say, that AMD Zen will not have "great performance" if you compare it against a comparable ARM design.
So no, AMD is no option at all if you are shooting for highest possible performance within the thermal and power limits of each device category.

 

[Jury]

Not sure that you understood what i wrote. I did say, that AMD Zen will not have "great performance" if you compare it against a comparable ARM design.
So no, AMD is no option at all if you are shooting for highest possible performance within the thermal and power limits of each device category.

No. Intel x86 performance is still slightly faster if there are on the same node with Apple SoC.

 

[Birkan]

Gruber's Daring Fireball blog is quoting Brendon Shanks hypothesis that it's going go be the A12Z SOC from the iPad Pro inside a Mac Mini or Apple TV chassis, which requires less engineering to create a developer-only box than a full MacBook. If so, I wonder if the A12Z can run at a higher thermal envelope then when wedged inside an iPad? Or if it's going to be a chip in a different package with something like a bigger heat spreader? Maybe real (non-dev) ARM Mac customers will have to wait till next year for some higher wattage variants of the A13 or A14.

Dalrymple's Loop blog might be passing along a whisper that Apple already has these systems working, ready to go, and that they are "fast".

The hints are out that Apple wants developers to really pay attention to the coming Keynote and Platform sessions.
[automerge]1592196957[/automerge]


But the fastest (known) CPU is on the Voyager 2 spacecraft.

I honestly do not think it will be any A12 series based SOC inside the so called development kit that Apple will provide. It totally makes sense to have a Apple TV / Mac Mini type device but I believe it will be at least some sort of A13 or newer series SOC since it has newer ARMv8.4 ISA on the CPU side and many improvements on the GPU side such as SIMD Group Instructions or ASTC HDR. Furthermore, using a newer ISA such as ARMv8.5 which would provide additional benefits such as memory tagging for improved security or ARMv8.6 which would provide improved Pointer Authentication. Implementing ARM SVE in addition to traditional NEON instructions could also help improve performance on Mac systems. Of course, it will basically come down to where Apple draws the line and accept as baseline for the future of ARM based Macs.

 

[appahappa]

No chance this being in a pro machine or desktop. Dedicated GPUs required x86 instruction sets, so how would an ARM play with an AMD GPU?

This is not true. You could use ARM CPUs and AMD GPUs together. GPUs don't use x86 instructions.

 

[chucker23n1]

Take the Surface Pro X as example. It is the highest performant 7W TDP Windows device. Even the recently released Intel Lakefield SoC, which is supposed to compete in the 7W space does not hold a candle against the SQ1 of the Surface Pro X performance wise.

Hmm, https://gadgetversus.com/processor/microsoft-sq1-vs-intel-core-i7-1065g7/ lists the chip as 9W. That would make it close to Ice Lake-NY's 10W, i.e. the MacBook Air.

What kind of data do you have on SQ1 perf?
[automerge]1592221883[/automerge]

I expect 20% improvement > than each intel part an apple arm is replacing.

I wonder if Apple stuck anything interesting in their hardware to make compatibility a little easier. One thing that comes to mind is the weird intel floating point (80 bits? It’s been awhile since i worked on x86 FP and drawing a blank, though I do remember it was a real pain after having worked on FP on powerpc and SPARC). Another is extra registers dedicated to emulation (both general purpose and things like stack pointer). If someone asked me to design an ARM cpu that could emulate x86 well, I can think of some things I could do with not-too-many transistors, particularly if I am already supporting heterogenous cores and I already have an SoC methodology which lets me build extra little cpus and stick them in the package.

One or two special X14 cores on the A14M to optimize x86 emulation sounds like an interesting approach, but I'm not sure it's something Apple wants to do in terms of engineering complexity. Maybe on the higher-end Macs?

 

[cmaier]

@cmaier - thank you for sharing.

Having played around with Assembler on various MCU architectures back in the day I can appreciate you comment about the RISC vs CISC advantage.

I have heard several times that Intel has (had?) a good manufacturing process but bad chip designers. Now the process improvement has stalled for 5 - 6 years and TSMC has the edge. Any root cause for Intel loosing the process edge and can it be recovered or will TSMC now be the leader for the future.

That I don’t know. Intel hasn’t been all that forthcoming with an explanation of what went wrong. But it’s been long enough that it’s hard to believe they will be able to not only catch up but exceed TSMC soon.


it’s also possible that tsmc hits a roadblock. But Intel led for a long time so maybe tsmc will too. One advantage Apple has is they can always just go with whoever is in the lead. Samsung, tsmc, globalfoundries, and yes even Intel would likely take apple’s money.

this is part of the reason AMD spun off its fabs. Being stuck to one supplier has some advantages, but also a lot of risks.
[automerge]1592227083[/automerge]

Hmm, https://gadgetversus.com/processor/microsoft-sq1-vs-intel-core-i7-1065g7/ lists the chip as 9W. That would make it close to Ice Lake-NY's 10W, i.e. the MacBook Air.

What kind of data do you have on SQ1 perf?
[automerge]1592221883[/automerge]


One or two special X14 cores on the A14M to optimize x86 emulation sounds like an interesting approach, but I'm not sure it's something Apple wants to do in terms of engineering complexity. Maybe on the higher-end Macs?

other than the fact that they couldn’t use the additional engineering work for anything other than their Mac line, it would probably be worth it for some of the things I could think of.

hell, at amd there were a couple occasions where we came very close to building those kinds of changes in for a couple big customers and the deals just didn’t get done. Extensions to the architecture just for running a certain kind of operating system or certain kind of software. Always a trade off in effort vs reward, but in apple’s case may be worth it, especially since arm is designed to allow all sorts of optional feature sets.

 

[alex2792]

no - AMD is the only company that has this license and intel fought them pretty hard (and lost) to limit them a long while back. they will never issue another license like AMD has again.

maybe apple should just buy AMD

The x86 license is not transferable, otherwise AMD would’ve been bought out a long time ago.

 

[Gerdi]

Hmm, https://gadgetversus.com/processor/microsoft-sq1-vs-intel-core-i7-1065g7/ lists the chip as 9W. That would make it close to Ice Lake-NY's 10W, i.e. the MacBook Air.

What kind of data do you have on SQ1 perf?

Not sure where the 9W comes from - maybe a typo. Microsoft says 7W.

Anyway if we compare Geekbech 5 under Windows:

Microsoft SQ1 (Surface Pro X) - 7W - 2900
Intel 1030NG7 (Macbook Air 2020) - 10W - 2600
Intel i5-L16L7 (Lakefield) - 7W - 1650

It is even worse for the Intel chips if we using some longer running benchmark, as with Geekbench the chip almost never throttles due to short duration, say 7-zip benchmark:

Microsoft SQ1 (Surface Pro X) - 7W - 17400MIPS
Intel 1030NG7 (Macbook Air 2020) - 10W - 10900MIPS

 

[chucker23n1]

Not sure where the 9W comes from - maybe a typo. Microsoft says 7W.

I see various sources claiming 7W (though whether that refers to something like a TDP equivalent is unclear), but… Microsoft's own site doesn't seem to mention this at all.

Anyway if we compare Geekbech 5 under Windows:

Microsoft SQ1 (Surface Pro X) - 7W - 2900
Intel 1030NG7 (Macbook Air 2020) - 10W - 2600
Intel i5-L16L7 (Lakefield) - 7W - 1650

So, eight Qualcomm Kryo 495 cores score 2900, and four Intel Sunny Cove cores score 2600? (Geekbench's browser actually lists 2909 for the Air i7.) Sounds like Intel is way ahead on that one?

 

[ignatius345]

The most stabilizing thing I have seen and appreciate is cloud storage tho. I don’t miss having to switch from the latest obsolete storage gizmos to another.

I'm with you on the utility of cloud storage, but complete physical backups have saved my ass so many times. And yes, that includes retrieving clean versions of files that got corrupted by being synced in cloud storage.