Apple CPUs "just margins off" desktop CPUs - Anandtech

[mikeymikec]

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.

Your opinion makes little sense if you don't back it up with something of substance; just because a processor is new, it doesn't mean it's the best available in any circumstances, nor does age affect CPU performance. Frankly I can't think of any obvious reasons why his post "makes no sense at all". StinkyPinky's response at least included a factual point as well as an obvious yet logical guess.

A common and important question with pretty much any solution in any setting is "does it scale well". For CPUs it's a common barrier for the most obvious reason that an architecture can only be clocked up so far (regardless of cooling mechanisms) before its inefficiency far outstrips the performance gain or it simply doesn't work.

 

[StinkyPinky]

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.

It's actually a very common thought that the A11 (and at a guess the A12) don't scale well. If it was just a matter of "making it a 35W tdp and slapping on a cooler" then logically Apple would have pursued that already. In case you haven't noticed Apple like to control as much of their ecosystem as possible

Not to say Apple won't indeed swap out their laptop/desktop cpu's with A13 cpu's in the near future, but like I said, it's a bold claim to suggest they can just ramp up with zero proof. Also, act with a bit more decorum next time.

 

[Gideon]

I agree that it's a bold claim but i certainly wouldn't rule it out. At least for fanless Macbooks and Airs.

After all may used the same argument against the Pentium M. How it had more IPC than the best desktop at the time (Athlon 64) but it won't ever scale. Then came Core 2, based on the same chip, and destroyed everything.

We'll have a better idea, once the new Ipad is released with higher TDP.

 

[darkswordsman17]

It means a certain figure that I don't want to publicly put in such a high profile article without having full confidence behind the comparison numbers on the desktop side until it's more thoroughly addressed in a dedicated article because everybody is going to quote it for months.

So it's a margin off.

So how is Anand these days?

 

[Spartak]

It's actually a very common thought that the A11 (and at a guess the A12) don't scale well. If it was just a matter of "making it a 35W tdp and slapping on a cooler" then logically Apple would have pursued that already. In case you haven't noticed Apple like to control as much of their ecosystem as possible

Not to say Apple won't indeed swap out their laptop/desktop cpu's with A13 cpu's in the near future, but like I said, it's a bold claim to suggest they can just ramp up with zero proof. Also, act with a bit more decorum next time.

Can you please support your claim that the A11 and A12 dont scale well? they have only been used in 3-4W TDP budgets so its completely pulled from thin air. it also betrays a limited insight on what scaling actually means. You seem to insinuate they are in their upper range limit already, which is quite a silly claim for a 3.6W chip. Read up on the various processes TSMC has for offer and the thermal/frequency ranges associated. It does mean a significant investment by Apple's process engineers but those are necessary miles to make, not fundamental problems they need to solve.

You also make it sound as if making a platform architecture switch is peanuts and use the lack of that till now to support your claim. Absurd. Also, see the previous point how you do need to invest significantly not just in the process itself but the entire system architecture including interconnects and the memory subsystem.

Finally, the whole premise of this thread was the conclusion by Andrei that the 3.6W A12 doesnt need scaling to compete with top of the line Intel CPU's. The 3.6W is already within striking distance of Intel's 95W CPU's for single core performance. Which is the only relevant metric for your scaling argument.

 

[Mopetar]

Isn’t Apple under some agreement with Intel for a few more years to use only Intel chips in their Mac computers?

Also, I don’t know if they have the volume to justify a separate line (or perhaps lines) of chips for their Macs, especially if they compete for wagers Apple needs for their mobile devices.

 

[Thala]

It's actually a very common thought that the A11 (and at a guess the A12) don't scale well. If it was just a matter of "making it a 35W tdp and slapping on a cooler" then logically Apple would have pursued that already.

You argument is flawed. Ever thought about the possibility, that there is no 35W TDP A-series chip, because Apple has no product for it yet?
In addition of course, it is not common though, because its a property of any CMOS logic, that frequency scales with voltage - this holds in particular if you are coming from a relatively low voltage as used in a mobile chip. It is also common knowledge, that frequency scales with lower threshold voltage - again low VT cells is something you trying avoid in a mobile chip.

 

[cytg111]

For the desktop its game over for x86 the day an arm chip pulls ahead in the next COD/BF title... Hell that's the day ill consider making the switch.

 

[TheGiant]

There is no deny that A12 is excellent and IPC wise probably on par with x86 best(that is CFL IPC), scaling up might be not that easy.
With x86 chips we are always talking about downscaling (CFL wasn't designed for 5W TDP with all its parts like AVX2 units, uncore etc).
But with Apple's speed of improvement I think we will read in one year - Apple announced iMac and MacBook with brand new the win A14 chip more powerful than...
it might not be that far away, considering we wont see icelake IPC until 2020, most probably so in fact Intel will push Apple to do such move with the inability to deliver

 

[Andrei.]

The 3.6W is already within striking distance of Intel's 95W CPU's for single core performance.

Can we stop the bullshit TDP comparison figures for Intel/desktops? Those 95W are when all cores are on. If I run two demanding threads on the A12 it'll also go up to 10W before it throttles.

 

[Thala]

Can we stop the bullshit TDP comparison figures for Intel/desktops? Those 95W are when all cores are on. If I run two demanding threads on the A12 it'll also go up to 10W before it throttles.

While i agree, that TDP should never be used to reason about efficiency - why don`t you include x86 CPUs into your energy comparisons?

 

[VirtualLarry]

considering we wont see icelake IPC until 2020, most probably so in fact Intel will push Apple to do such move with the inability to deliver

That's what (supposedly) got Apple to move off of PPC and Motorola as their supplier. Inability to deliver meaningful performance increases on a steady pace.

 

[jpiniero]

That's what (supposedly) got Apple to move off of PPC and Motorola as their supplier. Inability to deliver meaningful performance increases on a steady pace.

It was more Motorola's inability to deliver a sellable mobile chip.

 

[Charlie22911]

A12 is pretty thermal limited. The Slingshot test on my XS Max heats it up shockingly fast below the camera module.

I’m tempted to throw a peltier on the back of the phone just to see what it does

EDIT:
This is my best run with a 45w peltier:

I don’t seem to have any issues with power and crashes on my phone by the way...

 

[krumme]

For the desktop its game over for x86 the day an arm chip pulls ahead in the next COD/BF title... Hell that's the day ill consider making the switch.

I tend to agree there is far more truth to this than generally acknowledged.
As I said back when amd won the console deals it was the best that ever happened to Intel.
X86 is fragile.

When an Intel engineer visited Anand years back with a voltmeter and lectured about ohms law, we were led to believe that the x86 power myth was busted.
It was accompanied to the tune of 4b a year poured into atom.
We know where that went. Atom went bust. Anand ate an Apple.
To this day we can't get the power numbers and bm. Yeaa its difficult to adress but it's also explosive stuff.

 

[StinkyPinky]

Can you please support your claim that the A11 and A12 dont scale well? they have only been used in 3-4W TDP budgets so its completely pulled from thin air. it also betrays a limited insight on what scaling actually means. You seem to insinuate they are in their upper range limit already, which is quite a silly claim for a 3.6W chip. Read up on the various processes TSMC has for offer and the thermal/frequency ranges associated. It does mean a significant investment by Apple's process engineers but those are necessary miles to make, not fundamental problems they need to solve.

You also make it sound as if making a platform architecture switch is peanuts and use the lack of that till now to support your claim. Absurd. Also, see the previous point how you do need to invest significantly not just in the process itself but the entire system architecture including interconnects and the memory subsystem.

Finally, the whole premise of this thread was the conclusion by Andrei that the 3.6W A12 doesnt need scaling to compete with top of the line Intel CPU's. The 3.6W is already within striking distance of Intel's 95W CPU's for single core performance. Which is the only relevant metric for your scaling argument.

I was responding to the comment that a higher TDP A12 cpu would ""smash x86 processors completely if you package it like a desktop CPU". I am not the one that needs to support any claim, it is usually the person making the accusation in the first place (it wasn't you that made it, I am aware of that). I wasn't getting into the debate which architecture is better or which one has more of a future.

I am simply pointing out we know nothing about A-series CPU's and how they perform when given more juice to play with. It was also mentioned that if the A series cannot get over 3ghz, then Apple could just "add more cores" but we all know it isn't that easy. We only need to look at the threadripper cpu's that adding more cores isn't a magical solution to everything.

In short, we know *nothing* about how these cpu's will perform when given higher tdp. If it is anything like the x86, there will be diminishing returns.

 

[cytg111]

I tend to agree there is far more truth to this than generally acknowledged.
As I said back when amd won the console deals it was the best that ever happened to Intel.
X86 is fragile.

When an Intel engineer visited Anand years back with a voltmeter and lectured about ohms law, we were led to believe that the x86 power myth was busted.
It was accompanied to the tune of 4b a year poured into atom.
We know where that went. Atom went bust. Anand ate an Apple.
To this day we can't get the power numbers and bm. Yeaa its difficult to adress but it's also explosive stuff.

Yea, Microsoft needs to get win10 on this soc ASAP.

 

[krumme]

Yea, Microsoft needs to get win10 on this soc ASAP.

This. I hope every ms board member understands who their future customers is.

 

[krumme]

I was responding to the comment that a higher TDP A12 cpu would ""smash x86 processors completely if you package it like a desktop CPU". I am not the one that needs to support any claim, it is usually the person making the accusation in the first place (it wasn't you that made it, I am aware of that). I wasn't getting into the debate which architecture is better or which one has more of a future.

I am simply pointing out we know nothing about A-series CPU's and how they perform when given more juice to play with. It was also mentioned that if the A series cannot get over 3ghz, then Apple could just "add more cores" but we all know it isn't that easy. We only need to look at the threadripper cpu's that adding more cores isn't a magical solution to everything.

In short, we know *nothing* about how these cpu's will perform when given higher tdp. If it is anything like the x86, there will be diminishing returns.

Sure you can't eat a double whopper bacon and cheese as fast as a plain flat hamburger.
Still the double whopper constitutes the better meal if you can afford it.

 

[StinkyPinky]

Sure you can't eat a double whopper bacon and cheese as fast as a plain flat hamburger.
Still the double whopper constitutes the better meal if you can afford it.

Actually I had one of those yesterday, and ate it pretty damn fast. At least 4ghz.

 

[IntelUser2000]

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...

They won't, yet at the same time they do already.

The marketshare you are talking about is already achieved by many iPhones and iPads that stunted PC growth for a decade. It IS the "PC" that you are talking about for many people.

Competing for $200-$300 near thin client stuff and basic PCs is not really their bag.

Yes. Because they choose not to sell low cost, low margin devices. If they wanted, they could have, but they don't. It would result in dilution of the brand. Besides, they are doing perfectly well in terms of making money. They have 15% of the Smartphone market in terms of volume but take something like 80% of revenue. Yea, I bet you they are pretty happy about that.

In "PCs", I've read they have 80% marketshare of high end devices. So in the $1k+ range 80% of systems being sold are Apple? Yea, they've basically taken the PC in all the important segments.

By selling a device, rather than a component, Apple has an enormous advantage. Not only that, they only sell premium devices. Intel sells ~250 million chips a year. Apple sells about 200 million iPhones. However, Intel has an ASP of mere $100 range. iPhone ASPs are in the $700 range. Intel has to pull out all its stops to get 60% margins. Apple may have only 38% margins, but they bring in 6x the revenue.

Intel is also stuck with having to bifurcate their lines in an arbitrary fashion to keep their miserly $100 ASPs. So they have to make a little core that sells for much lower, despite the die sizes not being that much smaller, and a big core that sells for much more.

Apple, does not have to worry about that. They only have to keep 38% margin on a device-level. They could design Core-like chips without having to worry about the chip margin.

Samsung, is a combination of Intel and Apple. Samsung tries to use Apple strategy where they sell devices, and also Intel's where they make components. But, they divide their products into dozens and dozens of products that span from the cheapest, to the most expensive.

 

[tempestglen]

@Andrei.

Thanks for your article!

I don't know whether you are interested in time line.

1993----Daniel W. Dobberpuhl founded and directed DEC’s Palo Alto, California Design Center in 1993 where the StrongARM architecture was designed.

1993-1997----StrongARM was the best ARM chipset in the world.

1997----DEC agreed to sell StrongARM to Intel as part of a lawsuit settlement in 1997.[3] Intel used the StrongARM to replace their ailing line of RISC processors, the i860 and i960. When the semiconductor division of DEC was sold to Intel, many engineers from the Palo Alto design group moved to SiByte, a start-up company designing MIPS system-on-a-chip (SoC) products for the networking market.

1998-- After DEC defunct, Dobberpuhl co-founded SiByte, where as president he led the design of the SB1250 high performance MIPS system-on-a-chip processor.

2000----SiByte was bought by Broadcom. Dobberpuhl stayed until 2003 as vice president and general manager of the Broadcom broadband processor division.

2003----Dobberpuhl left Broadcom to found P.A. Semi, a fabless semiconductor company that designed the PWRficient family of Power Architecture processors.

2004----A gentleman's agreement between Dobberpuhl and Steve Jobs, PA6T was designed for Macintosh.

2005----PA6T, an excellent embedded chipset with rough same integer performance with Intel core, half float point performance, but very low power consumption.

2006----Steve Jobs betrayed Dobberpuhl by introducing Intel core to Macintosh platform, P.A.Semi business was in danger. "iPad" had been researched before iPhone project. Jobs wanted to use Atom but stopped by his team.

2008---- Apple acquired P. A. Semi. Steve Jobs made a plan for their own Microarchitecture ARM chipset.

2009----iPhone came out. Customized microarchitecture was still being researched. Dobberpuhl left apple.

2010----Apple acquired Intrinsity, another company with long history cooperation with Apple in 1990s when they designed chipset for Mac in the name of "Exponential Technology".

2012----A6 chipset came out.

2013----A7
2014----A8
2015----A9
2016----A10
2017----A11

2018----A12 outperforms Xeon IPC in SPECint 2006.

2019----A13, Will the successor of StrongARM/PA6T, land on Macintosh in the end? Welcome home! Anyway, ARM Holding was founded in November 1990 as Advanced RISC Machines Ltd and structured as a joint venture between Acorn Computers, Apple Computer (now Apple Inc.) and VLSI Technology.

https://en.wikipedia.org/wiki/Daniel_W._Dobberpuhl
https://en.wikipedia.org/wiki/StrongARM
https://en.wikipedia.org/wiki/P.A._Semi
https://www.realworldtech.com/pa-semi/
https://en.wikipedia.org/wiki/Intrinsity
https://en.wikipedia.org/wiki/Arm_Holdings

Obviously, Apple successful acquired talent engineers from DEC and other former PowerPC cooperators. Intel is not monopolist in CPU design field.

 

[naukkis]

A12 IPC seems to be well over 50% better in specint2006 than Skylake. It seems plain obviously that x86 time as performance cpu architecture starts to be over.

 

[stuff_me_good]

@Andrei.

Thanks for your article!

I don't know whether you are interested in time line.

1993----Daniel W. Dobberpuhl founded and directed DEC’s Palo Alto, California Design Center in 1993 where the StrongARM architecture was designed.

1993-1997----StrongARM was the best ARM chipset in the world.

1997----DEC agreed to sell StrongARM to Intel as part of a lawsuit settlement in 1997.[3] Intel used the StrongARM to replace their ailing line of RISC processors, the i860 and i960. When the semiconductor division of DEC was sold to Intel, many engineers from the Palo Alto design group moved to SiByte, a start-up company designing MIPS system-on-a-chip (SoC) products for the networking market.

1998-- After DEC defunct, Dobberpuhl co-founded SiByte, where as president he led the design of the SB1250 high performance MIPS system-on-a-chip processor.

2000----SiByte was bought by Broadcom. Dobberpuhl stayed until 2003 as vice president and general manager of the Broadcom broadband processor division.

2003----Dobberpuhl left Broadcom to found P.A. Semi, a fabless semiconductor company that designed the PWRficient family of Power Architecture processors.

2004----A gentleman's agreement between Dobberpuhl and Steve Jobs, PA6T was designed for Macintosh.

2005----PA6T, an excellent embedded chipset with rough same integer performance with Intel core, half float point performance, but very low power consumption.

2006----Steve Jobs betrayed Dobberpuhl by introducing Intel core to Macintosh platform, P.A.Semi business was in danger. "iPad" had been researched before iPhone project. Jobs wanted to use Atom but stopped by his team.

2008---- Apple acquired P. A. Semi. Steve Jobs made a plan for their own Microarchitecture ARM chipset.

2009----iPhone came out. Customized microarchitecture was still being researched. Dobberpuhl left apple.

2010----Apple acquired Intrinsity, another company with long history cooperation with Apple in 1990s when they designed chipset for Mac in the name of "Exponential Technology".

2012----A6 chipset came out.

2013----A7
2014----A8
2015----A9
2016----A10
2017----A11

2018----A12 outperforms Xeon IPC in SPECint 2006.

2019----A13, Will the successor of StrongARM/PA6T, land on Macintosh in the end? Welcome home! Anyway, ARM Holding was founded in November 1990 as Advanced RISC Machines Ltd and structured as a joint venture between Acorn Computers, Apple Computer (now Apple Inc.) and VLSI Technology.

https://en.wikipedia.org/wiki/Daniel_W._Dobberpuhl
https://en.wikipedia.org/wiki/StrongARM
https://en.wikipedia.org/wiki/P.A._Semi
https://www.realworldtech.com/pa-semi/
https://en.wikipedia.org/wiki/Intrinsity
https://en.wikipedia.org/wiki/Arm_Holdings

Obviously, Apple successful acquired talent engineers from DEC and other former PowerPC cooperators. Intel is not monopolist in CPU design field.

Interesting read and benchmark. Would be nice to know what the power consumption is during these benches?

 

[cytg111]

A12 IPC seems to be well over 50% better in specint2006 than Skylake. It seems plain obviously that x86 time as performance cpu architecture starts to be over.

Single threaded performance? That thing we thought was limited by whats physical possible? No more low hanging fruits?
I am still sceptical.