Apple CPUs "just margins off" desktop CPUs - Anandtech

[IntelUser2000]

It is quite an impressive result.

However, single thread comparisons are skewed in favor of low power parts. The Core i7 8559U at 15W should come within 10% of 8700K's ST result.

Desktop CPUs are all about having many of those cores, and vector instructions like AVX.

The article compared Xeon Platinum 8180's result. A 4.5W 7Y75 should rival that in single thread.

 

[jpiniero]

Also look at how huge the A12 is - 88MM2 for a dual core CPU with integrated graphics on the latest TSMC 7NM process.

Plus 4 small cores, the NPU, etc...

 

[USER8000]

It is quite an impressive result.

However, single thread comparisons are skewed in favor of low power parts. The Core i7 8559U at 15W should come within 10% of 8700K's ST result.

Desktop CPUs are all about having many of those cores, and vector instructions like AVX.

The article compared Xeon Platinum 8180's result. A 4.5W 7Y75 should rival that in single thread.

Plus the Apple parts are running their own OS with their own optimisations to get the max out of their hardware.

Its like comparing a console to a gaming PC. Yes,its all great,but when you are essentially comparing a 6.9 billion transistor dual core CPU(so basically think of a dual core Celeron) on the most dense and power efficient process node available,with CPUs packing massively more cores(and using less transistors) or those with significantly faster IGP sections,how huge will an 8 core A12 if had to run at least at 4GHZ,but on Global Foundries/Samsung/Intel 14NM?? Imagine how much more memory bandwidth it would need.

Also the article talks about wide designs - in Russia there were wide designs,like the Elbris ones which technically had very high IPC,but didn't scale very high in clockspeed.

Plus 4 small cores, the NPU, etc...

Intel and AMD also integrated ARM cores too,extensive chipset functions,especially if you look at Ryzen which is an SOC. All on less denser and less efficient processes.

They are relying on the tiny smaller cores,since they can't drop power of the larger ones sufficiently for normal low level tasks.

The fact is Apple is relying on new nodes much more than Intel or AMD is. What happens if the next nodes get drawn out?? I like to see how they can scale up performance if they can't throw billions more transistors at each generation.

The A11 had 4.3 billion transistors - the A12 6.9 billion. So A13 will be over 10 billion transistors on 5NM or 3NM?? What happens if they can't rely on cutting edge nodes to drop power per transistor enough??

So now scale that up to 4GHZ~5GHZ operation(more transistors),a more complex memory controller(more transistors),more cores(more transistors),things like AVX(more transistors) and so on.Oh,lets add a proper interconnect to enable scaling for more cores??

Something like Infinity Fabric or Mesh,since these also consume a ton of power,as you need scaleability. Now add that to the A12.

How big will a proper desktop class A12 be?? How much more power will it be eating??

Each of these phone launches is so world changing we might as well start replace our minds with them soon - smartphone indeed.

Do they look awesome for their form factor? Sure,but its like comparing the latest technowonder electric 2 seater town car to a 4X4 jeep. Just because it does some things much better,its not like suddenly it will be replacing a jeep.

Its the same problem with existing X86 cores,they are designed for higher performance,higher TDP markets,so trying to drop down to the areas ARM has excelled in,is the same problem.

Plus I am sure Apple with try and move to its own cores,but for many years it has not been at the forefront of performance in its X86 systems anyway.

But what is the likelihood the Apple desktop replacement ARM based cores will be not small chips at all even on cutting edge manufacturing.

 

[IntelUser2000]

Plus the Apple parts are running their own OS with their own optimisations to get the max out of their hardware.

Its like comparing a console to a gaming PC.

I'm not sure that comparison is apt from a performance perspective, but maybe the control allows them to iterate faster, as they know what's coming on the software side. Intel has to wait for vendors.

Intel is also executing pretty horribly. If they had Icelake last year, it'd have been faster and used less power compared to Amber/Whiskey.

Amberlake may be pretty comparable in performance to the A12 for ST. What Apple's execution advantages afford them is the ability to put that performance in a Smartphone. That is a substantial advantage, but not out of the world as "performing like a desktop chip" quotes tell you.

Intel and AMD also integrated ARM cores too,extensive chipset functions,especially if you look at Ryzen which is an SOC. All on less denser and less efficient processes.

Bit off topic, but Intel is moving rapidly away from them. The XMM 7560 LTE modem in the latest iPhone uses an x86 core. Earlier Linkedin reports had it pegged as an Atom core. I'd bet its like a 400MHz Silvermont. Their server chipset uses an embedded Quark core since Skylake. Their modem division uses Atom cores too.

The fact that they can use Silvermont on an LTE modem while being competitive on a power usage front is big - at least they know how to make efficient cores so it can be proliferated further. They should have done this long time ago. Point of x86 everywhere is so they get to learn from it and apply it on their main cores.

 

[MagnusTheBrewer]

I'm not in the right tax bracket for Apple to have any impact on me.

 

[jpiniero]

So now scale that up to 4GHZ~5GHZ operation(more transistors)

As it is, the A12 is very competitive with top of the line desktop Core/Zen parts as it is in a lot of the browser benchmarks and Geekbench. I imagine it's very competitive in a lot of tasks. Is 4+ Ghz really needed?

I'm actually kind of interested as to how they are connecting the big and small cores; and if an A12X comes along with more CPU cores.

 

[USER8000]

I'm not sure that comparison is apt from a performance perspective, but maybe the control allows them to iterate faster, as they know what's coming on the software side. Intel has to wait for vendors.

Intel is also executing pretty horribly. If they had Icelake last year, it'd have been faster and used less power compared to Amber/Whiskey.

Amberlake may be pretty comparable in performance to the A12 for ST. What Apple's execution advantages afford them is the ability to put that performance in a Smartphone. That is a substantial advantage, but not out of the world as "performing like a desktop chip" quotes tell you.

Don't get me wrong is great performance in such a small TDP,but if you look throw the design,ie,only dual cores but very wide ones,you can tell its been developed for mobile operations,at lower clockspeeds on effiency orientated lower leakage nodes. Looking at the "desktop" cores from AMD and Intel,they probably give up potential core IPC just so they can clock higher,scale to more cores,etc. Even the process nodes they use which are deemed as "high performance" probably mean more leakage,etc.

Can I see a desktop equivalent ARM based core - sure,but I don't think it will just as easy or as small a chip as people think it will be,even if X86 is a shoe-horned thing.

In fact I would rather see more coverage of RISC-V and it is a shame when Linus pays more attention to that than mainstream tech websites:

He even looked at the first production RISC-V chip.

Bit off topic, but Intel is moving rapidly away from them. The XMM 7560 LTE modem in the latest iPhone uses an x86 core. Earlier Linkedin reports had it pegged as an Atom core. I'd bet its like a 400MHz Silvermont. Their server chipset uses an embedded Quark core since Skylake. Their modem division uses Atom cores too.

The fact that they can use Silvermont on an LTE modem while being competitive on a power usage front is big - at least they know how to make efficient cores so it can be proliferated further. They should have done this long time ago. Point of x86 everywhere is so they get to learn from it and apply it on their main cores.

I didn't know that.

As it is, the A12 is very competitive with top of the line desktop Core/Zen parts as it is in a lot of the browser benchmarks and Geekbench. I imagine it's very competitive in a lot of tasks. Is 4+ Ghz really needed?

I'm actually kind of interested as to how they are connecting the big and small cores; and if an A12X comes along with more CPU cores.

They are kind of synthetic benchmarks,and again very few and very wide cores,which need probably a lot of resources to keep them at peak utilisation,and again all done under a closed OS with tweaks. This is why they needed to rejig the memory subsystem. Wide cores,also means they want to keep at a lower clockspeed range - ever thought what pushing higher clockspeeds means(IPC given up for longer pipelines and use of leakier high performance processes),but then very wide designs have their own problems too. Its not the first time designers have tried to go wider and lower clocked.

Will a supercomputer have an A12 core?? Last time I checked IBM is supplying the CPUs for ORNL.You might as well compare a console to a gaming desktop. One runs games very efficiently with what it has,and the other brute forces it.

As usual at EVERY phone launch,more and more hype. Also for web browsing even a cheapo Android phone or old computer is fine.Most Android phones sold are not higher end devices worldwide and Android makes up 77% of the smartphone OS share. Things like web browsing and word processing have not needed much power for a very longtime.Its why smartphones are reaching 3 year lifespans,and laptops/desktops are being kept longer and longer.

People are instantly calling Intel,AMD doomed,just because Apple made a 7 billion transistor dual core,on the most efficient node the world has. All great,but they are just throwing transistors at the problem.

Is it going to run something like Fallout 4 fine,which is very single core bottlenecked,with 100s of NPCs,massive draw calls,etc?? Be able to do software 4K transcoding on only two cores,etc??

This is all hype TBH,as Apple,Samsung need to find new ways to increase phone prices to increase margins,as more and more people are holding off upgrading. What happened to the PC market is now happening to the phone market.

In reality what would be more useful is not all this pointless increases in processing power for a phone,but phones which can have have functions like GPS,data,etc used a lot but last a week between charges.

Things like trickle solar recharging through the screen which I believe Apple and Samsung looked at,but nothing has come out of it.

I would imagine for most phone users,their battery running out at an awkward time is more of an issue than X amount more performance,when ordering some trainers using the Amazon app,or watching Taylor Swift,etc on YouTube.

Dropping power draw more and more makes more sense,in terms of battery liffespan,as more charge/discharge cycles is what destroys batteries - some of the phones and tablets with liquid cooling,is really a step backwards. They should not need it.

Battery tech is what is the limiting factor nowadays and wearing down the batteries quicker with largish power draws,to push benchmark figures is great as companies can have built-in obsolescence selling more phones(worse with Android). Most of the batteries are dumped as its difficult to recycle them - it is the hidden cost of smartphones and tablets.

 

[Thala]

Bit off topic, but Intel is moving rapidly away from them. The XMM 7560 LTE modem in the latest iPhone uses an x86 core. Earlier Linkedin reports had it pegged as an Atom core. I'd bet its like a 400MHz Silvermont. Their server chipset uses an embedded Quark core since Skylake. Their modem division uses Atom cores too.

The fact that they can use Silvermont on an LTE modem while being competitive on a power usage front is big - at least they know how to make efficient cores so it can be proliferated further. They should have done this long time ago. Point of x86 everywhere is so they get to learn from it and apply it on their main cores.

First you have no idea how power efficient the XMM7560 is and second, XMM7660 has no x86 cores anymore.

 

[jpiniero]

I guess it's more of a question whether Apple can dump Intel and not look back. So far the answer is yes, provided that Apple can deliver on something like EMIB.

 

[IntelUser2000]

First you have no idea how power efficient the XMM7560 is and second, XMM7660 has no x86 cores anymore.

Source on the XMM 7660?

And yes, we do have an idea its pretty efficient as the overall battery life is quite decent. If the idle power is high as with their previous efforts, the overall battery life would tank. You also sound a bit angry.

Things like trickle solar recharging through the screen which I believe Apple and Samsung looked at,but nothing has come out of it.

If you played around with Solar, you'd see why its not used. Making the screen transparent means you'd lose most of the light energy. That's why solar cells are dark. Solar cells are also current sources making circuitry needed much more complex than a battery to utilize it properly.

 

[Arkaign]

Great posts Mr 8000. Yes, scaling to different markets and OS/use cases is pretty difficult. Just look at how AMD and Intel fared with their lowest power SKUs. X86 didn't seem to go too great in the downward direction and tablet/phone situation. Neither have we seen Qualcomm or Apple offer products that seem to be competent replacements for something like a Ryzen or Core i7 in full workstation clothes.

 

[USER8000]

If you played around with Solar, you'd see why its not used. Making the screen transparent means you'd lose most of the light energy. That's why solar cells are dark. Solar cells are also current sources making circuitry needed much more complex than a battery to utilize it properly.

Well I would rather phones actually did move away from one silly oneupmanship on how fast the Amazon app works,on whether Taylor Swift videos run better,and moved more towards pushing battery life. All this trying to beat Intel,AMD,IBM,doesn't end up giving us something we didn't already have,and I would argue software is one of the main reasons we see current very powerful PC hardware not performing as well as it should. We all know why Apple is doing this - to drop production costs further by removing the middle man,just like when they screwed Imagination Technologies over.

Great posts Mr 8000. Yes, scaling to different markets and OS/use cases is pretty difficult. Just look at how AMD and Intel fared with their lowest power SKUs. X86 didn't seem to go too great in the downward direction and tablet/phone situation. Neither have we seen Qualcomm or Apple offer products that seem to be competent replacements for something like a Ryzen or Core i7 in full workstation clothes.

Yes,I am impressed at what they can do,but come on,every phone launch is the doom of all other tech humanity ever created.

Its the overhype at every phone launch.Even Linus had to make an iPhone video:

It could be I am just getting a bit over-exposed to the incessant fawning over each phone launch,on tech sites,photography sites,everywhere and then its all of a sudden normal PCs are doomed(!!),cameras are doomed(!!),gaming PCs and consoles are doomed(!!),books are doomed(!!),hifi is doomed(!!),TVs are doomed(!!),etc. Everyone,this £1000 phone will do all of that better(!!),even though £1000 is a ridiculous cost for a cheap to make phone,which will only last a few years,but it will save you more money(!!),promise(!!).

Jack of all trades, master of none.

 

[jpiniero]

^^^

They did manage to kill the headphone jack.

 

[USER8000]

^^^

They did manage to kill the headphone jack.

They had murder on their mind obviously - people need to not be worried about Skynet,it will be all the "AIs" in smartphones connecting through social media,and then dooming us!!

 

[name99]

Why? Apple is never going sell these chips to anyone else and their PC hardware business isn’t that big (although I bet it’s quite profitable for intel).

And Intel has a win on the modem.

(a) Never is a long time.

(b) Apple is a large server CPU purchaser. And there is no reason this core (or more precisely the core based on it optimized for desktop) couldn't also form the basis for a very nice server chip. The only missing piece I see is a NoC that can handle "many" (say 48 or more) cores, and you think that's beyond Apple?

(c) The term "desktop" or "PC" is somewhat vague. It's assumed to mean what it means today, a very specific bundle of UI (large screen + keyboard + mouse), extensibility (at least one, preferably multiple ports like USB that connect to HW), and OSs that have a certain feel and a certain set of expectations (whether Mac, Windows, or Linux). But it's unclear that these HAVE to all be that way. In particular, it's possible that an OS that's much more like iOS (more locked down, more difficult to customize, but also more secure and more difficult to break) attached to the HW of large screen, kbd, pointing device would be exceptionally popular, enough so to achieve (in time) iOS levels of acceptance rather than Mac levels of acceptance.

It's possible to quibble about the details of this (especially if you insist on being a moron, rather than accepting the larger point). eg
- Will there be windows? I don't see why not. Windows aren't a central feature of iOS because of screen area concerns, not because they are insecure or difficult for people to understand.

- Will there be connections to external HW? This is trickier. There is tremendous value in external hardware (particularly storage devices, but also eg ethernet, or external GPUs) but this hardware is definitely one of the weak links in the quality of experience for PCs generally (ie both Macs and Windows). There is just too much lousy HW out there, and the misbehavior of the OS in the face of lousy hardware remains too opaque. One way Apple might handle this is a rather more decentralized OS which delegates every interaction with external HW to a standalone core, perhaps running a standalone OS (or a less drastic version of the same sort of thing, running via hypervisor a primary OS and an IO OS, with the primary OS seeing only virtual devices). Point is, regardless of the details, to ensure that
(a) HW that fails, crashes, etc simply cannot bring down the machine, no matter what
(b) EVERY HW failure (including unexpected timing failures) is audited and can be reported.
Failing HW can either simply be rejected or, worst case, can be prominently brought to the user's attention.
(As opposed to now where, for example, a common reason that the experience sucks on older PCs --- Mac and Windows --- is a slowly failing hard drive that struggles along but, every so often, takes 30 seconds to perform some task as it tries to read some bad sector, or tries to sync to a track that's no longer well demarcated.0

Point is: Apple could create something that feels like a PC so far as the things MOST people care about (big screen(s), windows, pointing device, external hard drive) but with the security and safety of iOS rather than macOS. Obviously some people will be very unhappy, but Apple's not interested in selling to the build-you-own PC, compile your own Linux crowd; just like, say, Toyota, sells cars that average people want to buy, not cars that are designed for you to modify them as you wish.
And if Apple capture iOS levels of the PC market (which I think is quite feasible given the model I'm describing, though it may take 15 years to get there because of the slower turnaround of PCs compared to phones) that is a BIG problem for Intel...

 

[mattiasnyc]

In servers, it might hurt AMD more since conservative buyers would stick with intel, adventurous buyers would buy Apple instead of Ryzen.

I really don't see how they'd get a large volume of server chips sold. As a prospective buyer you'd have to look at longevity and upgrade paths and that basically means the commitment of the supplier to that product and platform.

Ask yourself what Apple's current reputation is for providing professional platforms that conforms to industry standards and then maintaining and developing those products and solutions over many years time.

I'd trust Apple with a server farm as far as I could throw one.

Apple can't even properly cater to the professional creative community in my opinion.

 

[Arkaign]

Software is the biggest problem facing Apple on the business front. Mac is almost an afterthought for current corporate focus as it is. Their margins are staggering on iPhones and good on iPads, but a fair bit less on the Mac business. There's a lot of software that doesn't show up there. QuickBooks Enterprise ended with 2016 :/

With the growth of phone as the defacto device most lean on, I don't really see a situation where Apple is really interested in pushing more for a market that is already filtering towards NUC style devices anyway. Competing for $200-$300 near thin client stuff and basic PCs is not really their bag.

 

[LightningZ71]

Why do you think it would ever be that cheap? Nothing in their laptop portfolio is that cheap. In teardowns of their low end laptops, it's staggering how little there is in them.

 

[beginner99]

If I were Intel, I'd be worried.

No. Apple doesn't have the laptop&desktop volume to create multiple chips and with each model multiple options. They would need at least 1 more chip/SOC to serve the mac pro line or simply ditch it. In reality the would probably need 3 dies. Not worth it.

Add to that the software issues that will ensue. Simply not worth it.

If Apple had a server chip that could bring them in $$$, do you really think they wouldn't tool up to serve that market? Its another high margin product that would fit very well into Apple's philosophy.

Apple servers end-users (consumers!) and part of their philosophy is the design of their products which people pay absurd amounts for. Server market is something completely different and design matters jack.

Also look at how huge the A12 is - 88MM2 for a dual core CPU with integrated graphics on the latest TSMC 7NM process. Look at the 6 core Intel Core i7 CPUs running upto 5GHZ - they are 149MM on a less dense process and probably having far less transistors.

Exactly. Apple can throw huge amounts of silicon at the problem because the chips will sell in only couple devices for very high price. Intel and AMD must design their cores to work at high frequency and in different configs (dual, quad, hexa,...) with some of the products ending up in dirty cheap products. hence you can't just throw endless silicon at the problem.

If you data is correct an A12 core is roughly 3.5 times bigger compared to Intel Core while being on a lower / denser node. One would expect all that area to have some effect on performance. And who knows how apple cheats in benchmarking with power limits off or other stuff...

 

[WelshBloke]

...(c) The term "desktop" or "PC" is somewhat vague. It's assumed to mean what it means today, a very specific bundle of UI (large screen + keyboard + mouse), extensibility (at least one, preferably multiple ports like USB that connect to HW), and OSs that have a certain feel and a certain set of expectations (whether Mac, Windows, or Linux). But it's unclear that these HAVE to all be that way. In particular, it's possible that an OS that's much more like iOS (more locked down, more difficult to customize, but also more secure and more difficult to break) attached to the HW of large screen, kbd, pointing device would be exceptionally popular, enough so to achieve (in time) iOS levels of acceptance rather than Mac levels of acceptance.

It's possible to quibble about the details of this (especially if you insist on being a moron, rather than accepting the larger point). eg
- Will there be windows? I don't see why not. Windows aren't a central feature of iOS because of screen area concerns, not because they are insecure or difficult for people to understand.

- Will there be connections to external HW? This is trickier. There is tremendous value in external hardware (particularly storage devices, but also eg ethernet, or external GPUs) but this hardware is definitely one of the weak links in the quality of experience for PCs generally (ie both Macs and Windows). There is just too much lousy HW out there, and the misbehavior of the OS in the face of lousy hardware remains too opaque. One way Apple might handle this is a rather more decentralized OS which delegates every interaction with external HW to a standalone core, perhaps running a standalone OS (or a less drastic version of the same sort of thing, running via hypervisor a primary OS and an IO OS, with the primary OS seeing only virtual devices). Point is, regardless of the details, to ensure that
(a) HW that fails, crashes, etc simply cannot bring down the machine, no matter what
(b) EVERY HW failure (including unexpected timing failures) is audited and can be reported.
Failing HW can either simply be rejected or, worst case, can be prominently brought to the user's attention.
(As opposed to now where, for example, a common reason that the experience sucks on older PCs --- Mac and Windows --- is a slowly failing hard drive that struggles along but, every so often, takes 30 seconds to perform some task as it tries to read some bad sector, or tries to sync to a track that's no longer well demarcated.0

Point is: Apple could create something that feels like a PC so far as the things MOST people care about (big screen(s), windows, pointing device, external hard drive) but with the security and safety of iOS rather than macOS. Obviously some people will be very unhappy, but Apple's not interested in selling to the build-you-own PC, compile your own Linux crowd; just like, say, Toyota, sells cars that average people want to buy, not cars that are designed for you to modify them as you wish.
And if Apple capture iOS levels of the PC market (which I think is quite feasible given the model I'm describing, though it may take 15 years to get there because of the slower turnaround of PCs compared to phones) that is a BIG problem for Intel...

You're describing a Chromebook.

 

[Spartak]

We are two full pages into this thread and nobody mentions the elephant in the room?

____ oo
|||| J


I don't think only AMD will use TSMC's high performance 7nm process.

 

[Charlie22911]

I believe the reason Intel should be worried is because it demonstrates the performance potential of the architecture.
No longer is ARM limited to low power, low performance embedded devices as it was when I was coming up. Apple has shown in the past a willingness to change the status quo, even if it stings. It doesn’t seem unreasonable to expect them to replace intel in their PCs, I’d argue they are already moving that direction with their security SOCs in the newer devices.
More ARM based devices means more developer support, which makes a transition away from an architure wielded by a industry bully not only more likely, but inevitable IMO.

 

[Spartak]

It is quite an impressive result.

However, single thread comparisons are skewed in favor of low power parts. The Core i7 8559U at 15W should come within 10% of 8700K's ST result.

what the hell are you talking about? The 8559U@4.5GHz gets a SC geekbench score of ~5050, the 8086K@5GHz gets about ~5750. That's apples to apples a disproportionate advantage for the higher clocked part (both have the same amount of cache per core). Probably the 8559U is not maintaining its max turbo throughout every test.

People are throwing way too much dust in the air here. The IPC of the A12 is already vastly higher compared to the Skylake family, so much that SC performance of a 3-4W part is equal to 20W Intel parts on half the frequency. They might not hit 5 or even 4GHz on their theoretical ARM desktop Macs, but a 8-24Core CPU range @ just 3 GHz turbo would likely already outperform anything Intel has to offer. I'd expect them to hit at least 4.

 

[StinkyPinky]

A12 would already smash x86 processors completely if you package it like a desktop CPU, give it a big heatsink like a desktop CPU, and give it a similar amount of power compared to a desktop CPU.

Bold claim. If that were true, why would they even bother with Intel at all. It's likely the A12 doesn't scale up well, much in the same way Intel's cpu's don't scale down well.

 

[Spartak]

Bold claim. If that were true, why would they even bother with Intel at all. It's likely the A12 doesn't scale up well, much in the same way Intel's cpu's don't scale down well.

the A12 is just out. the desktop line hasnt been updated in ages. Your post makes no sense for all the obvious reasons I wont spell out to you. Just think it through a bit before posting.