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

[Eug]

Rumor is A10X is on TSMC 10nm, so could actually be smaller/similarly sized.

I haven't seen that rumour. Where did you see that, and is the source at all reliable?

 

[Arachnotronic]

I haven't seen that rumour. Where did you see that, and is the source at all reliable?

KGI.

 

[Intervenator]

Apple mentioned that the headphone jack was taking too much space due to its aged design, and that they needed that room for additional hardware. Was this to make room for the big.LITTLE since they need additional room for the additional hardware? Or, if I recall correctly from the presentation, they had a haptic feedback engine near the bottom of the phone. Did this take its place?

Also, what are the odds that Intel does its own vision of big.LITTLE for their mobile and ultra mobile offerings? It seems that Apple switching to big.LITTLE has put the final nail in the coffin to the debate on the benefits of this tech.

Regardless, I look forwards to a teardown of the internal design of the 7. Its amazing how every year and without fail smartphone makers (most notably Apple) add more new components while at the same time condensing / obsoleting old ones. Its something that I always look forwards to because this is the true driver of innovation in modern day mobile.

 

[stingerman]

In any case, you can't really say Apple makes the best microprocessors when their chips are a measly 4W TDP. .

"Measly 4W" - That's what is amazing, so much performance at that level. That's where the talent is. We can only imagine what Apple's design team can do with a relaxed TDP Spec.

But it helps for them that they're designing chips that the only go in >$600 products.

No, they design all sorts of chips at all sorts of price points . Maybe you meant flagship processors, not chips. Keep in mind, most computers sell for < $600.

 

[stingerman]

Was this to make room for the big.LITTLE since they need additional room for the additional hardware? Or, if I recall correctly from the presentation, they had a haptic feedback engine near the bottom of the phone. Did this take its place?

This is not a big.LITTLE design and it would not directly have been affected by the jack. Apple's design appears to use a unified cache between all four cores and they added a fast hardware switch to allocate a thread to the appropriate core. It just might be that all four cores can run simultaneously. With threads that don't contain any conditional branching or complex floating point operations going to the more efficient cores. That makes a lot more sense to me.

 

[stingerman]

What's interesting to me is Apple moving traditionally kernel level decision making onto the processor. So can Fusion also mean the hardwiring of key parts of the Mach/BSD kernel onto the A10 (not to mention the runtime engine.) Something I've always expected to happen. So I suspect, we haven't heard half the A10 story yet.

What are they waiting for? No doubt the switch away from Intel. If the A10 is greatly accelerating the Apple OS by hardwiring key components, then we are looking at a major boost in performance along with a great reduction in power.

 

[Andrei.]

What's interesting to me is Apple moving traditionally kernel level decision making onto the processor. So can Fusion also mean the hardwiring of key parts of the Mach/BSD kernel onto the A10 (not to mention the runtime engine.) Something I've always expected to happen. So I suspect, we haven't heard half the A10 story yet.

This is so far-fetched that I just don't see it being possible. You basically need the hardware to be aware of everything that is traditionally part of the kernel scheduler. How would this controller be aware of threads? You still need information from the kernel at which point it's arguable what the advantage to do this in hardware is?

I still think the controller is a hardware DVFS governor which at the same time decides discrete switching (only either one of the two is active) between pairs of little and big cores. Again it seems the OS only sees 2 cores here so for me this makes the most sense.

 

[Andrei.]

Double post

 

[Eug]

This is not a big.LITTLE design and it would not directly have been affected by the jack. Apple's design appears to use a unified cache between all four cores and they added a fast hardware switch to allocate a thread to the appropriate core. It just might be that all four cores can run simultaneously. With threads that don't contain any conditional branching or complex floating point operations going to the more efficient cores. That makes a lot more sense to me.

If that Geekbench result is legit, it should be noted that Geekbench is still seeing two cores.

 

[stingerman]

If that Geekbench result is legit, it should be noted that Geekbench is still seeing two cores.

Should be interesting to see more details unfold

This is so far-fetched that I just don't see it being possible. You basically need the hardware to be aware of everything that is traditionally part of the kernel scheduler. How would this controller be aware of threads? You still need information from the kernel at which point it's arguable what the advantage to do this in hardware is?

I still think the controller is a hardware DVFS governor which at the same time decides discrete switching (only either one of the two is active) between pairs of little and big cores. Again it seems the OS only sees 2 cores here so for me this makes the most sense.

It's always faster to write something in software, but any software can be hardened if you have the resources and budget. Think of all the operations that go on in the camera sensor, separate from the CPU cores. It's an advantage you have when you can control the whole stack and directly influence the hardware design.

 

[witeken]

"Measly 4W" - That's what is amazing, so much performance at that level.

Woah, when Apple makes 4W high-end chip, it gets lauded as the highest engineering achievement of mankind.
When Intel makes 4W high-end chip with 50% higher peak performance and performance per watt than A10, no one gives a [censored].

Fanboyism much (also looking at those 3 likes).

 

[teejee]

Woah, when Apple makes 4W high-end chip, it gets lauded as the highest engineering achievement of mankind.
When Intel makes 4W high-end chip with 50% higher peak performance and performance per watt than A10, no one gives a [censored].

Fanboyism much (also looking at those 3 likes).

Try to put Intels 4W high-end chip in a 140 gram smartphone. Well, actually no need to try that, I can tell you it won't work well.



Sent from my LG-D855 using Tapatalk

 

[witeken]

it won't work well.

Only because it isn't designed for it.

http://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

Llama Mountain, 7.2mm thick (BDW-Y).
iPhone 7, 7.1mm thick.

 

[swilli89]

This is a really lame response. Apple didn't force anybody to buy the 16GB version, it was mearly an offering. As you said, it was the minimum.

I don't know how offering options to consumers makes a company "the worst".

The old "make people pay more for what should be standard" line in the guise of "offering options". Apple should have damn well been at 32gb for a few years now but we both know they would have lost some 64gb sales.

And yes your next counter is probably something along the lines of "Capitalism, maximizing profits etc"... You can save that. My response was to the guy claiming everything should be included in a "flagship" phone in the first place.

 

[Arachnotronic]

Only because it isn't designed for it.

http://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

Llama Mountain, 7.2mm thick (BDW-Y).
iPhone 7, 7.1mm thick.

Thickness in a vacuum isn't that helpful. Overall volume is a much more meaningful metric for what you are trying to prove.

 

[Arachnotronic]

The old "make people pay more for what should be standard" line in the guise of "offering options". Apple should have damn well been at 32gb for a few years now but we both know they would have lost some 64gb sales.

And yes your next counter is probably something along the lines of "Capitalism, maximizing profits etc"... You can save that. My response was to the guy claiming everything should be included in a "flagship" phone in the first place.

I should have been more precise in my language.

A premium phone should have a premium processor.

 

[Arachnotronic]

The old "make people pay more for what should be standard" line in the guise of "offering options". Apple should have damn well been at 32gb for a few years now but we both know they would have lost some 64gb sales.

And yes your next counter is probably something along the lines of "Capitalism, maximizing profits etc"... You can save that. My response was to the guy claiming everything should be included in a "flagship" phone in the first place.

Apple is there now, at 32GB base. Does this make it more likely that you're going to buy the next iPhone, or were you never a potential Apple customer anyway?

 

[SAAA]

Just a theory now that we have A10 details: could the first leaked result be showing the two slower cores clocks? ~400MHz sounds about right for low-power cores and 2GB ram could be for iPhone7 vs 7+.

 

[Eug]

Just a theory now that we have A10 details: could the first leaked result be showing the two slower cores clocks? ~400MHz sounds about right for low-power cores and 2GB ram could be for iPhone7 vs 7+.

Interesting point. Ironically though, the 400 MHz result had higher scores.

But yeah, does anyone know how Geekbench estimates the clock speed?

BTW, back in the old days I had a Cube running a 1.7 GHz G4 Sonnet upgrade, but OS X said my clockspeed was 0 MHz.

 

[lopri]

Those scores indicate more throttling than A9. The MT score is not even double that of the ST score.

 

[Eug]

Those scores indicate more throttling than A9. The MT score is not even double that of the ST score.

Why would it be double? A9 isn't even close to double:

A10: 5363 / 3233 (1.66X)
A9: 2407 / 4046 (1.68X)
A8: 1476 / 2487 (1.68X)
A7: 1209 / 2035 (1.68X)

Shockingly consistent actually.

BTW, there are now 4 different iPhone 7 models listed there at Geekbench.

iPhone9,1 (2 GB): 5592 / 3382
iPhone9,2 (3 GB): 5552 / 3418
iPhone9,3 (2 GB): 5495 / 3379
iPhone9,4 (3 GB): 5363 / 3233

Legit or not I don't know. If not, then they're some pretty reasonable fakes.

 

[asendra]

Those scores indicate more throttling than A9. The MT score is not even double that of the ST score.

Considering that doubling the score is the theorical maximum they could do, I find that fine.
The OS won't see the four cores at the same time

 

[witeken]

Apple is there now, at 32GB base. Does this make it more likely that you're going to buy the next iPhone, or were you never a potential Apple customer anyway?

TBH, 32GB is still ridiculous. I'm not just pointing at Apple. The savings on NAND by these phone manufacturers is crazy. Not to say the laptop market is any better, both have their problems. But phones should come with at least 128GB, just like laptops. In the post just above that, you even said:

A premium phone should have a premium processor.

Premium phone should also have premium storagy. Enough said.

Speaking of NAND, the costs of those have also slown down considerably, and 3D NAND isn't really helping much either, so far.

 

[witeken]

BTW, there are now 4 different iPhone 7 models listed there at Geekbench.

iPhone9,1 (2 GB): 5592 / 3382
iPhone9,2 (3 GB): 5552 / 3418
iPhone9,3 (2 GB): 5495 / 3379
iPhone9,4 (3 GB): 5363 / 3233

Legit or not I don't know. If not, then they're some pretty reasonable fakes.

What scores do Intel processors have (I'm not really all that much into benchmarking)?

 

[Nothingness]

What scores do Intel processors have (I'm not really all that much into benchmarking)?

You can look here: http://browser.primatelabs.com/processor-benchmarks