Apple A10 Fusion is ** Quad-core big.LITTLE **

[witeken]

A10X should give a preview of what's to come with A11. I expect A10X to clock the CPUs at significantly higher frequencies, and I expect that with A11, Apple will be able to bring both a frequency improvement as well as a "tock-like" improvement in perf/MHz.

Apple's chip team is insane. And I mean this in all of the good senses of that word.

What I would give them credit for is their consistency. No Qualcomm we'll-use-off-the-shelf-ARM-this-year-because-we-have-nothing-in-our-pipeline-and-our-own-arch-next-year stuff. Just the old school 40% improvement per year, even if we can actually increase clock speed by much more but-hey-we-still-need-some-headroom-for-next-year.

But you can also see it in their R&D. Developing chips has ballooned their development costs. (Or is all of that going to the iCar?)

But don't keep dreaming. Physics will stand in Apple's way.

BTW, if you see an exponential curve into the $10B like this, doubling in 2 years, you wonder if it's gonna level off or if Apple planning to be at $40B R&D in 2020 ?

 

[Arachnotronic]

What I would give them credit for is their consistency. No Qualcomm we'll-use-off-the-shelf-ARM-this-year-because-we-have-nothing-in-our-pipeline-and-our-own-arch-next-year stuff. Just the old school 40% improvement per year, even if we can actually increase clock speed by much more but-hey-we-still-need-some-headroom-for-next-year.

But you can also see it in their R&D. Developing chips has ballooned their development costs. (Or is all of that going to the iCar?)

But don't keep dreaming. Physics will stand in Apple's way.

BTW, if you see an exponential curve into the $10B like this, doubling in 2 years, you wonder if it's gonna level off or if Apple planning to be at $40B R&D in 2020 ?

Never bet against the desire for profits coupled with human ingenuity.

 

[witeken]

Never bet against the desire for profits coupled with human ingenuity.

I don't think Apple's engineers are at all concerned about Apple's financials.

 

[Eug]

What I would give them credit for is their consistency. No Qualcomm we'll-use-off-the-shelf-ARM-this-year-because-we-have-nothing-in-our-pipeline-and-our-own-arch-next-year stuff. Just the old school 40% improvement per year, even if we can actually increase clock speed by much more but-hey-we-still-need-some-headroom-for-next-year.

But you can also see it in their R&D. Developing chips has ballooned their development costs. (Or is all of that going to the iCar?)

But don't keep dreaming. Physics will stand in Apple's way.

BTW, if you see an exponential curve into the $10B like this, doubling in 2 years, you wonder if it's gonna level off or if Apple planning to be at $40B R&D in 2020 ?

Those are catch-up-with-Intel expenditures. It doesn't seem like any of the other chip makers were willing to put in that much R&D money into their designs but Apple has and brought ARM to the forefront now.

The question I have though is whether or not the other ARM makers are going to be able to catch up with Apple.

 

[StrangerGuy]

The question I have though is whether or not the other ARM makers are going to be able to catch up with Apple.

Forget the A10, it would be a tall order for them to match the A9 ST in 2017 when ARM themselves more-or-less confirms that a 2.8GHz A73 will only be 30% faster at most than current 2.5GHz A72 which itself is already very competitive with Kyro / M1. The funny part is it looks like ARM's own cores are now better designed than their Android custom core partners.

 

[dark zero]

And now... It happened. Seems that ARM Hurricane is on works... And it includes support with MacOS. This is now an interesting escenario. A new brand one.

 

[Andrei.]

Forget the A10, it would be a tall order for them to match the A9 ST in 2017 when ARM themselves more-or-less confirms that a 2.8GHz A73 will only be 30% faster at most than current 2.5GHz A72 which itself is already very competitive with Kyro / M1. The funny part is it looks like ARM's own cores are now better designed than their Android custom core partners.

The µarch after next year's will be interesting to see how ARM keeps up, I guess it could be a convergence point between the two design philosophies. Samsung will also have something new at that time.

 

[Arachnotronic]

The µarch after next year's will be interesting to see how ARM keeps up, I guess it could be a convergence point between the two design philosophies. Samsung will also have something new at that time.

Is the next Samsung SoC just going to use Mongoose cores in 10nm?

 

[krumme]

A73 is less wide than prior cores and tuned for efficiency. It looks to me like lower cost and different segment than Apples solution. Do we have any estimates of cpu mm2 size including L2 for A73 vs. Apple A9?

 

[Andrei.]

A73 is less wide than prior cores and tuned for efficiency. It looks to me like lower cost and different segment than Apples solution. Do we have any estimates of cpu mm2 size including L2 for A73 vs. Apple A9?

Right. ARM targets a lot of segments with A73, Apple just has to worry about 1-2 SoCs. Don't have any 1:1 numbers but extrapolating ARM's 5mm² estimate for a quad 2MB L2 on 10FF and assuming a 1.46x scaling back to 16FF that's still more than half the size of Apple's 2 cores + 3MB L2.

Is the next Samsung SoC just going to use Mongoose cores in 10nm?

More or less, supposedly a revised Mongoose at very high clocks. The next big µarch change is the year after.

 

[scannall]

And now... It happened. Seems that ARM Hurricane is on works... And it includes support with MacOS. This is now an interesting escenario. A new brand one.

I wouldn't read too much into that. They have always maintained concurrent builds of MacOS.

 

[StrangerGuy]

More or less, supposedly a revised Mongoose at very high clocks. The next big µarch change is the year after.

They might as well just go with the A73 with ~16% better IPC than 8890/A72; I really doubt the clock gains will be worth the trouble.

 

[Andrei.]

They might as well just go with the A73 with ~16% better IPC than 8890/A72; I really doubt the clock gains will be worth the trouble.

Where'd you get 16% from? A73 is supposedly same IPC as A72 in SPEC.

 

[StrangerGuy]

Where'd you get 16% from? A73 is supposedly same IPC as A72 in SPEC.

I simply took Kirin 955 GB4 ST scores, added +30% to it, then divide it by 2.8GHz, in accordance to ARM's projections.

 

[Andrei.]

I simply took Kirin 955 GB4 ST scores, added +30% to it, then divide it by 2.8GHz, in accordance to ARM's projections.

That makes zero sense. ARM never made any GB4 projections plus they were clear on the IPC differences to A72: http://www.anandtech.com/show/10347/arm-cortex-a73-artemis-unveiled/3

 

[StrangerGuy]

That makes zero sense. ARM never made any GB4 projections plus they were clear on the IPC differences to A72: http://www.anandtech.com/show/10347/arm-cortex-a73-artemis-unveiled/3

https://www.arm.com/products/processors/cortex-a/cortex-a73-processor.php

"Cortex-A73 delivers 30% more performance than the Cortex-A72 current devices in a mobile power envelope."

"Implemented at up to 2.8GHz"

 

[Andrei.]

https://www.arm.com/products/processors/cortex-a/cortex-a73-processor.php

"Cortex-A73 delivers 30% more performance than the Cortex-A72 current devices in a mobile power envelope."

"Implemented at up to 2.8GHz"

Those figures are their estimate from both µarch as well as process at the time the µarch is supposed to be mostly used on. Apples and oranges.

 

[krumme]

I kind of follow what Strangerguy main point is.
Good for Samsung, Qualcomm and perhaps even Apple that Samsung dont intend to use a 10nm A73 A35/A53, for the next Samsung S8 in ½ a year alongside their own project. It whould show how meaningless their own strategy is. I recon the idea of a beefed up DSP for eg. 4k stabilized video and GPU to drive VR, but who needs more than A73 cpu perf levels? - it will deliver it at low production cost and at the same time delivering what the consumers needs most, battery life and far less throtling that have hindered especially A15 and A57.
Look how insanely popular A53 is. Even midrange is populated by this miniscule cpu. And battery life plays a big part here also. Its in order processing at the best.
Now we get a less wide high performance variant with the same benefits as A53. Its about time. I think it will be a home run.

 

[stingerman]

Most of the Android phones sold do not contain high-end ARM processors. This is a problem for Qualcomm and Samsung. Look at the Apple R&D expenditure posted earlier.

Each new process node has been getting more and more expensive leaving Apple in the best position to be the volume first mover. Apple's lead is looking like 1.5 years going on to three years in the next 6 months

 

[Qwertilot]

Well it isn't such a problem when the not high end stuff is getting really very good now.

 

[Arachnotronic]

Most of the Android phones sold do not contain high-end ARM processors. This is a problem for Qualcomm and Samsung. Look at the Apple R&D expenditure posted earlier.

Each new process node has been getting more and more expensive and Apple in the best position to be the volume first mover. Apple's lead is looking like 1.5 years going on to three years in the next 6 months

Apple is the clear leader in this market. It built several world class processor teams and they are just building these new processor cores like there's no tomorrow.

Apple is the premium mobile/computing device company, and it shows in its processor designs. People can try to downplay Apple, but the results really speak for themselves.

Can't wait to see how much they can squeeze out of that Hurricane core in A10X before they move on to the next core in A11. A11 should be a "tock" (A7 and A9 have been bigger jumps in per-clock performance/architecture than A8 and A10 were).

Just got my iPhone 7 Plus a little while ago, but I already can't wait for the A11/iPhone 8...

 

[StrangerGuy]

I kind of follow what Strangerguy main point is.
Good for Samsung, Qualcomm and perhaps even Apple that Samsung dont intend to use a 10nm A73 A35/A53, for the next Samsung S8 in ½ a year alongside their own project. It whould show how meaningless their own strategy is. I recon the idea of a beefed up DSP for eg. 4k stabilized video and GPU to drive VR, but who needs more than A73 cpu perf levels? - it will deliver it at low production cost and at the same time delivering what the consumers needs most, battery life and far less throtling that have hindered especially A15 and A57.
Look how insanely popular A53 is. Even midrange is populated by this miniscule cpu. And battery life plays a big part here also. Its in order processing at the best.
Now we get a less wide high performance variant with the same benefits as A53. Its about time. I think it will be a home run.

Qualcomm's own 28nm SD650/652 already makes their premium 14nm Kyro look like a bad joke after taking into account on both their ST IPC, process node and cost. With non-Samsung Android high-end sales slumping further and further I don't expect their custom cores to survive past 2017.

 

[krumme]

Yeaa. I dont think it really matter for qcom anyway. Their dsp, gpu and especially modem is what makes them interesting. To me it seems like an idiotic use of ressources and a bad priority.
Looking at the lukewarm reception of ip7 one have to wonder if apple shouldnt care more for the core product. But i guess they have another need for all that expensive cpu grunt comming up.

 

[Phynaz]

I don't think Apple's engineers are at all concerned about Apple's financials.

They should be if they want their bonus pay.

 

[Arachnotronic]

They should be if they want their bonus pay.

And job security. When the financials go south, that's when management likes to implement these things called "cost cutting measures" and "restructuring actions."

Plenty of ex-Intel, ex-AMD, and ex-Qualcomm employees (all three cos have laid off substantial numbers of people over the last five years due to unsatisfactory financial performance) can tell you all about it.