Apple A12 & A12X *** Now A12Z as well *** Now in a Mac mini

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Alpha translation worked very well because Digital cached the application's binary translation to the hard drive (and I was running 10K SCSI drives). If Rosetta 2 does this (Macs will have more storage space), then performance will be great. SSDs FTW!

Edit: Apple could add some silicon based functional unit that improves translation as well, since they design their own chips.


I would assume every JIT these days caches its output to speed things up, but Apple is going one better with Rosetta 2. It is doing static translation and converting binaries from x64 to ARMv8 at install time. For the binaries where this can be done (it won't be possible for everything) the performance should be hard to distinguish from native code.
 
  • Like
Reactions: Tlh97

Eug

Lifer
Mar 11, 2000
23,583
996
126
Of course Apple is still selling x86 Macs and will support them for probably 5-7 years after they quit selling them so those people need not be in any hurry to adopt ARM.
I wouldn't necessarily count on that. It depends on what you mean by "support" though.

The last PowerPC Macs were sold in 2006.

The first version of Mac OS X that didn't support PowerPC came out in 2009, just three years later. That was Snow Leopard.

If Apple stops selling Intel Macs by 2022, I suspect that if Apple is being aggressive, by 2025 Apple's newest version of macOS may not support Intel Macs anymore. Apple will still support those Macs through security updates, but that would be for older versions of macOS.

So, possible timeline:

2022: All Intel Macs discontinued.
2025: Newest macOS release does not support Intel Macs.
2027: Last security update for Intel Macs released.

That's fine by me though, since my Intel Macs are from 2017 and I don't plan on upgrading them any time soon. I hope to be able to use them until 2025 or so. I have an iMac 27" Core i5-7600 with SSD, and a MacBook 12" Core m3-7Y32. I'm fine with these machines as they both have hardware h.265 decode (even if it's not DRM'd h.265 decode), and both have sufficient RAM. (The iMac is 24 GB but is RAM upgradable, and the MacBook has 16 GB.) By 2025, any software I would be using that still supports macOS would have already been updated to support Arm, so my transition from Intel to Arm would be easy.

IOW, the timing works out perfectly for me. I have recent Intel machines that will be updated for years after the Arm transition, but which are old enough to warrant upgrades to Arm machines when macOS stops supporting Intel.

It could be that the timeline is too aggressive, but nonetheless the typical 5-7 years shouldn't be expected.
 
  • Like
Reactions: name99

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
That's 3 mask sets, but not all are new every year. I think they will need to do this to get the right performance for each segment.

I would think macbook air and macbook could get away with the same SOC as ipad pro. Then for the macbook pro, imac etc they will need a bigger one. The real issue it the mac pro which will yet need an ever better one just for mac pro. I can't see how that makes any economical sense or I greatly underestimate mac pro sales?
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
I would think macbook air and macbook could get away with the same SOC as ipad pro. Then for the macbook pro, imac etc they will need a bigger one. The real issue it the mac pro which will yet need an ever better one just for mac pro. I can't see how that makes any economical sense or I greatly underestimate mac pro sales?

They can use their own processors on datacenters.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
I would think macbook air and macbook could get away with the same SOC as ipad pro. Then for the macbook pro, imac etc they will need a bigger one. The real issue it the mac pro which will yet need an ever better one just for mac pro. I can't see how that makes any economical sense or I greatly underestimate mac pro sales?

They could get away with using the iPad Pro SoC though it would need some stuff added to it, like support for ethernet, Thunderbolt and so forth but that's not a big hurdle. It doesn't sound like this is what they're doing though.

They don't necessarily need a different SoC for the Mac Pro - there's no reason to make a huge SoC with a lot of cores like Intel makes their CPUs. I think they will make smaller chips that work together as 'chiplets' like AMD does, and if appropriately designed those chiplets would also be capable of working alone in smaller Macs.

Not saying they will do that, if they want the best possible performance for the Mac Pro the chiplets that go in it will be a separate design since unleashed from a laptop's power budget they'd be able to target a higher clock rate. There are some process 'knobs' that can be adjusted to trade density and power for faster transistor switching (i.e. higher frequencies) and it would make sense to unleash the beast on the Mac Pro / iMac Pro to make sure they put the performance of the x86 based Power Mac well in the rear view mirror.

As far as cost, I thought I saw somewhere once that Apple sells over a million Mac Pro & iMac Pro per year. But even if it was only a half that, and Apple somehow spent $250 million a year for design/mask costs specific only to the Mac Pro / iMac Pro that amortizes to $500 per unit. Sounds like a lot until you consider they probably pay 2-4x (depending on model) that much for the Xeons going in those Macs now!

They would have to spend an extraordinary amount on Mac Pro specific CPU design costs for it to not save money over what they pay Intel currently. The per chip variable cost (assuming they use chiplets) is tiny, only a couple hundred bucks at most. They've got a LOT of room to play with if they are basing what they invest in the Mac Pro's new CPU on the Mac Pro's current cost structure.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,571
136
Let's also not forget, that nothing stops Apple to also outsourcing some of it's niche ARM designs, if needed, at least initially (when designing a custom SKU for limited market doesn't make sense due to extremely low volume).

For instance, if they decide that they really need a Rendering Mac Pro with a throughput oriented monster SKU and their available A14 chips don't quite cut it for that workload. They might just offer some limited models with Altra Quicksilver Q64-26 (125W 64 cores) or something similar. The lower-end versions would still be their chips.

Or they might just licence build their own ARM-neoverse implementation (with minimal customizations and also 64+ cores).

Not that they have to, but there are some stop-gap options available for such niche-markets.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Apparently 'Appple has also outlined strict guidelines for developers who will be in possession of the ARM-powered Mac mini. They cannot run any benchmarks on it, and they aren’t allowed to disassemble it.'

So we'll have to rely exclusively on leaks for another half year.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
I would think macbook air and macbook could get away with the same SOC as ipad pro. Then for the macbook pro, imac etc they will need a bigger one. The real issue it the mac pro which will yet need an ever better one just for mac pro. I can't see how that makes any economical sense or I greatly underestimate mac pro sales?
Nah.

I'm thinking that the MacBooks/MacBook Airs could actually just run A14, the same SoC as the iPhone. Yes seriously.

I expect A14 to be roughly 15-20% faster than A13. If you're OK with Geekbenching, that would mean a Geekbench 5 score of about 4000, which actually puts it around the Geekbench performance of a 2018 MacBook Pro Core i7-8559U. I don't know how the memory support, etc. and differences in macOS vs iOS factor into this, but nonetheless even if a significant slowdown from these factors is considered, this type of performance is more than sufficient for a MacBook/MacBook Air. Note that this is with iPhone clock speeds. If Apple decides to go back to fanless MacBooks, then the iPhone SoC would work well. However, if they stick with the fan in the MacBook Air, they could even increase the clock speed a bit vs. the iPhones.

Similarly, for the MacBook Pro, A14X will be sufficient, up to at least the mid tier. The CPU Geekbench 5 speed will be probably in the 5500 ballpark. Also, the GPU of even the 2 year-old A12X is already faster than most Intel GPUs.

I'm not sure what Arm chip they had for the demos, but it has been said that the Rosetta'd Tomb Raider looked like it was running at 1080p with mid-quality settings and sustained 30 fps. In contrast, with these settings, Intel UHD G1 can't get into the double digits for fps.


Yes, for top end 16 MacBook Pros and for top end iMacs, they will need something more, but overall, I suspect most MacBook Pros and iMacs would actually be running the same (class of) chips as the iPad Pros. I find it rather telling that the Mac mini dev kit is just an A12Z, which is effectively 2 year-old technology, and they mated 16 GB to it just fine too.

As for Thunderbolt, I wouldn't be completely surprised if A14 & A14X have USB 4 support built-in, with Thunderbolt 3 support. That might be a bit optimistic though. It'd be less surprising if USB 4 doesn't show up until A15X in 2021/2022. That doesn't stop Apple from using an outboard Thunderbolt chipset in 2020 though.

As for the Mac Pro, that is a boutique Mac. Apple can afford to create a boutique chip for their boutique Mac, and still save money vs. what they pay to Intel for those Xeons.
 
Last edited:
  • Like
Reactions: teejee and Tlh97

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
As for the Mac Pro, that is a boutique Mac. Apple can afford to create a boutique chip for their boutique Mac, and still save money vs. what they pay to Intel for those Xeons.
Sales are probably too low for it too be worth the engineering and mask sets. If their higher end SoCs can be ganged together with some chiplet and connection fabric, then that would be the way to go.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Let's also not forget, that nothing stops Apple to also outsourcing some of it's niche ARM designs, if needed, at least initially (when designing a custom SKU for limited market doesn't make sense due to extremely low volume).

For instance, if they decide that they really need a Rendering Mac Pro with a throughput oriented monster SKU and their available A14 chips don't quite cut it for that workload. They might just offer some limited models with Altra Quicksilver Q64-26 (125W 64 cores) or something similar. The lower-end versions would still be their chips.

Or they might just licence build their own ARM-neoverse implementation (with minimal customizations and also 64+ cores).

Not that they have to, but there are some stop-gap options available for such niche-markets.

Not a chance in hell. The whole point of going to ARM was so they could control the features, schedule, etc. They aren't going to trust some little fly by night company that isn't guaranteed to even exist in a few years, nor are they going to license someone else's lesser performing core.
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
Sales are probably too low for it too be worth the engineering and mask sets. If their higher end SoCs can be ganged together with some chiplet and connection fabric, then that would be the way to go.

Honestly, I feel like Apple has tried to burn the high end desktop bridge so hard for so long I find it hard to believe they are really going to take it seriously.

The older Mac towers were great, expandable, attractive, stable, fast.

Then the trash can came... and sat. A seemingly promised update schedule turned out to be... nothing?

The latest Mac Pro? I mean, vibes of the older models, great launch stats, etc. But near as I can tell, no updates since, and I would guess now no more updates for it's lifecycle?

I feel like you'd need to be a little gullible to buy into that type of life cycle again.

In two years I imagine there will be some pretty ludicrous Epyc & Intel based systems to choose from that are going to be that much more of different beast than what is available now.

I guess it won't really surprise me if Apple has another kinda cool Mac Pro Arm launch with some stitched together solution that is novel/interesting and then they proceed to forget it exists for another five years or so.
 

zinfamous

No Lifer
Jul 12, 2006
110,511
29,096
146
The translation should be considered not much more than a crutch to get you along until the app gets converted.

OK cool, thanks. so pretty much a software thing that isn't going to limit first gen hardware...outside the not-yet-determined issue of whether or not common MS stuff will have something that works on these new Macs.

....one thing I forgot, do I assume that this is also true with OS and their likely updates? I'm assuming that any issues with OS updates past the first gen ARM hardware could be pretty easily patched if need-be, as this seems to be common with Windows anyway
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I guess it won't really surprise me if Apple has another kinda cool Mac Pro Arm launch with some stitched together solution that is novel/interesting and then they proceed to forget it exists for another five years or so.

They could just introduce a higher TDP version of the one in laptops and make a super tiny trash can, the type you see in modern offices.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Sales are probably too low for it too be worth the engineering and mask sets. If their higher end SoCs can be ganged together with some chiplet and connection fabric, then that would be the way to go.

I agree but they have to pay mostly the same engineering costs whether they use a 'chiplet' solution or a monolithic core - you have a fabric connecting all the cores either way whether it is all on chip or communicates between chiplets. They have to pay the same mask set costs for either the monolithic core or the chiplets unless the chiplets are designed so they can also operate standalone in smaller Macs.

If they aren't going to design SoCs that say go into a Macbook Pro as an 8 core CPU/GPU/etc. and also have fabric functionality that is unused in the Macbook Pro which allows that same chip to connect to others as chiplets and function as a 32 or 64 core CPU in the Mac Pro then they probably don't save a lot of money with the chiplet solution. The risk to the big monolithic core is yield, though given how much they pay Intel now for Xeons they could probably afford an absolutely horrible yield and still come out ahead - though the other risk to poor yield is schedule. If they have a shortage of working Mac Pro SoCs it would cause shipment delays. We've seen that more than a few times with GPU vendors when they have yield issues with their huge dies.
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
I agree but they have to pay mostly the same engineering costs whether they use a 'chiplet' solution or a monolithic core - you have a fabric connecting all the cores either way whether it is all on chip or communicates between chiplets. They have to pay the same mask set costs for either the monolithic core or the chiplets unless the chiplets are designed so they can also operate standalone in smaller Macs.

If they aren't going to design SoCs that say go into a Macbook Pro as an 8 core CPU/GPU/etc. and also have fabric functionality that is unused in the Macbook Pro which allows that same chip to connect to others as chiplets and function as a 32 or 64 core CPU in the Mac Pro then they probably don't save a lot of money with the chiplet solution. The risk to the big monolithic core is yield, though given how much they pay Intel now for Xeons they could probably afford an absolutely horrible yield and still come out ahead - though the other risk to poor yield is schedule. If they have a shortage of working Mac Pro SoCs it would cause shipment delays. We've seen that more than a few times with GPU vendors when they have yield issues with their huge dies.

Seems to me like the biggest risk is wasting the time of engineers who should be making the chips that go into tens of millions of devices 1% better at something vs trying to engineer a solution that might impact what, hundreds of thousands of devices? The tradeoffs are massive.

As a shareholder I want iDevices to be more profitable. First, second and third priorities. The rest is somewhere down the list.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
Remember, Apple/Mac has that blasted T2 chip that can become the "T3" if it needs to. This already handles hard disk I/O, security enclave services, and has an encoding/decoding engine and audio DSP. Apple seems more than willing to provide for the integration of a second chip that can provide flexible I/O services and accelerate certain functions. So, what isn't provided by the base A series processor that will most likely be in those Macs can be offloaded as needed into another piece of silicon that doesn't have to be anywhere near as advanced as the processor, and can persist for multiple years.

As for the higher end Mac Pros, I just don't see them going forward with massive, specialized processors. Instead, I see them just living with a modest respin of the same architecture as the rest of the processors, but instead of being big.LITTLE, I see them as having just 8 large cores instead. What will make them "special" is their ability to house expansion devices, like video cards, etc. IT may just be multiple Thunderbolt ports, or it could be internal PCIe slots, but definitely gobs of storage. I feel that Apple will be shifting its focus on the Mac Pro devices to providing a solid, seamless front end to their cloud computing products. They have been hiring a lot of Cloud Computing talent recently, which would cover growth in their cloud services. Their biggest issue there would be Gigantic video files from creatives. However, that can be easily addressed by using high end semi-custom GPUs from AMD as they have been as most of those functions are offloaded to GPUs anyway. That can be local easily enough without having a mammoth XEON under the hood.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
The latest Mac Pro? I mean, vibes of the older models, great launch stats, etc. But near as I can tell, no updates since, and I would guess now no more updates for it's lifecycle?

Might be another two years before Intel has any Xeon-W to theoretically upgrade to. Maybe Intel will release some minor spec bumps to Cascade Lake-W which Apple could use.
 
  • Like
Reactions: blckgrffn

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
The latest Mac Pro? I mean, vibes of the older models, great launch stats, etc. But near as I can tell, no updates since, and I would guess now no more updates for it's lifecycle?

No updates since? It was released six months ago, how fast are you expecting updates to come???
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
No updates since? It was released six months ago, how fast are you expecting updates to come???

My bad. Announced in April 2018, shown in June 2019 and finally shipping December of 2019 seems to have warped my perception of how long it has been available.

Given the flurry of CPU activity in the meantime I guess it also feels like longer. 28 max cores seems quaint ;) I seemed to also have forgotten that despite a massive spec sheet it is only 1S capable, or am I reading that wrong now?

Ditto for the lack of CPU updates @jpiniero. Your point makes it even more likely that the current chassis is at it's evolutionary peak now. If you are going to buy an Intel based Mac Pro, might as well go all out right now, it won't get any sweeter or more relevant...

Also, when I consider it more, is just another example of the reason they are moving away from Intel.

My quick googling to re-educate myself reveals that you can now get that gear saddle for your Mac Pro that you were lusting after. Real cow hide.

 
Last edited:

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
The latest Mac Pro? I mean, vibes of the older models, great launch stats, etc. But near as I can tell, no updates since, and I would guess now no more updates for it's lifecycle?

The latest Mac Pro is 6 months old. You have not seen a CPU upgrade for Intel has not released a CPU update in the last 6 months that is using the "server" parts. Intel Cascade Lake was launched in April '19.

Now Intel just announced Cooper Lake literally 4 days ago, it is still a Skylake architecture (so no CPU performance increase for the same mhz) but it uses a different socket so I do not see the benefit of it unless you need more than 28 cores. (Someone will come in here and correct me and say why Cooper Lake is superior to Cascade Lake, but I just do not see it.)
 
  • Like
Reactions: blckgrffn

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
Now Intel just announced Cooper Lake literally 4 days ago, it is still a Skylake architecture (so no CPU performance increase for the same mhz) but it uses a different socket so I do not see the benefit of it unless you need more than 28 cores. (Someone will come in here and correct me and say why Cooper Lake is superior to Cascade Lake, but I just do not see it.)

It's basically Cascade Lake with bfloat added. Still 28 cores max.
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
The latest Mac Pro is 6 months old. You have not seen a CPU upgrade for Intel has not released a CPU update in the last 6 months that is using the "server" parts. Intel Cascade Lake was launched in April '19.

Now Intel just announced Cooper Lake literally 4 days ago, it is still a Skylake architecture (so no CPU performance increase for the same mhz) but it uses a different socket so I do not see the benefit of it unless you need more than 28 cores. (Someone will come in here and correct me and say why Cooper Lake is superior to Cascade Lake, but I just do not see it.)

Right, I caught that in my post above... but the point to take away here is if Apple stayed with Intel on this platform I suppose it would have been ~2022 before there would have been reason to refresh anyway. Maybe we would have seen some RDNA2 GPUs but even that seems like an unlikely effort at this point in time.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Right, I caught that in my post above... but the point to take away here is if Apple stayed with Intel on this platform I suppose it would have been ~2022 before there would have been reason to refresh anyway. Maybe we would have seen some RDNA2 GPUs but even that seems like an unlikely effort at this point in time.

I suspect the "Apple should upgrade the Mac Pro" crowd wants them to put AMD in there. That would be a worthwhile performance bump, but it would be a lot of work - and definitely not worth it to do all the work to support AMD when they are leaving x86 behind anyway.
 
  • Like
Reactions: blckgrffn

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
given how much they pay Intel now for Xeons
We don't know how much they pay. I'm willing to bet it's less than 20% of list price but really we will never know.

A SOC for Mac pro isn't worth it directly. Only reason to make it (and break even at best) is to have an actual powerful content creation machine for your entire platform but then the type of people using macs, an imac would probably suffice already for that.