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

[ElFenix]

P.S. Apple was working on this in theory at least at some point. This is from their patent filing:

motorola atrix already did this. no further patents should be issuable.

 

[Eug]

motorola atrix already did this. no further patents should be issuable.

Well regardless, here is Apple’s patent filing for it.

http://appft.uspto.gov/netacgi/nph-...and+pd/3/23/2017&RS=(CCL/1/1+AND+PD/20170323)

 

[frozentundra123456]

If Intel pulled a %15 increase people would be drooling all over the place. ;-)

As will the AMD crowd x10 if zen 2 increases performance by 15%.

 

[Kenmitch]

As will the AMD crowd x10 if zen 2 increases performance by 15%.

AMD achieving parity with Intel in single core would be a monumental achievement....Why wouldn't they after a decade of despair?

Poke pun at it if you find it amusing....Internet cpu forum meltdown....I can't wait to read the silly comments when it happens.

 

[Arachnotronic]

Apple is running circles around Intel. This is too funny.

 

[Mopetar]

If Intel pulled a %15 increase people would be drooling all over the place. ;-)

That's because Intel hasn't had a 15% jump in a long while, so it would be big for them. Apple typically sees more than 15% so it feels like a disappointment. Everything is relative.

motorola atrix already did this. no further patents should be issuable.

Didn't the Atrix just dock the phone with the laptop? It looks like the Apple patent wants to use the phone as a trackpad. Also, patents don't cover general ideas, but implementations. You can have two patented solutions that accomplish the exact same thing, but if the implementations are different, it's possible for both patents to be valid with neither infringing on each other or preventing someone else from coming up with a third, independent implementation.

That aside, I don't think Apple would ever release something like that as they change the form factor of their phones way too often. Unless you get the computer and the phone at the same time, they probably won't be compatible and upgrading one, means that you lose the ability to use it with the other.

 

[dark zero]

Yes. There is a Huawei though, also 6.9 billion transistors.

Kirin 980 which is Cortex A76. Also 2 big and 4 LITTLE cores.

Nope. Kirin 980 are:
- 2x A76 @ 2.8 Ghz
- 2x A76 @ 2.2 Ghz
- 4x A55 @ 1.8 Ghz

Are octacores

 

[Kenmitch]

That aside, I don't think Apple would ever release something like that as they change the form factor of their phones way too often. Unless you get the computer and the phone at the same time, they probably won't be compatible and upgrading one, means that you lose the ability to use it with the other.

It doesn't really look like anything special as a phone can be used as a mouse already. They're just kind of doing the Apple bundle the technology thing and claim it as the second coming of the trackpad.

 

[darkswordsman17]

That's because Intel hasn't had a 15% jump in a long while, so it would be big for them. Apple typically sees more than 15% so it feels like a disappointment. Everything is relative.



Didn't the Atrix just dock the phone with the laptop? It looks like the Apple patent wants to use the phone as a trackpad. Also, patents don't cover general ideas, but implementations. You can have two patented solutions that accomplish the exact same thing, but if the implementations are different, it's possible for both patents to be valid with neither infringing on each other or preventing someone else from coming up with a third, independent implementation.

That aside, I don't think Apple would ever release something like that as they change the form factor of their phones way too often. Unless you get the computer and the phone at the same time, they probably won't be compatible and upgrading one, means that you lose the ability to use it with the other.

I disagree that they do that some ridiculously high amount (its less than basically almost all the other phone manufacturers). They had the iPhone 6 form for 4 years (6, 6S, 7, 8), and it could arguably be 5 as the new ones I don't think are drastically different. They could always sell a bumper that lets the newest slimmer phone fit the cavity (and make the cavity for the biggest sized device at the time, where they could keep similar overall size but increase screen by reducing bezels and the like). And its not like they couldn't standardize if they were planning on such a product. Heck they could just make the cutout ridiculous sized so that it could support cases and other things as well (where if its a 3rd party case it'd come with its own bumper probably).

I think you actually just perfectly explained why Apple would do that as well. Imagine if they could get people to pony up for a new $1000 phone and also a new $500-1000 laptop shell every couple of years. For some people this would actually make it a bit more affordable since they could maybe the shell every 3-4 years (like when they'd upgrade their laptop), while buying the phone every other year. And it'd offer a more compelling upgrade probably.

 

[Eug]

Nope. Kirin 980 are:
- 2x A76 @ 2.8 Ghz
- 2x A76 @ 2.2 Ghz
- 4x A55 @ 1.8 Ghz

Are octacores

Thx. I’ve fixed my post.

 

[NTMBK]

EDIT: N/m.

 

[DrMrLordX]

Do know. Could be the 11x is targeted toward the next iPad. That, or Apple just skipped it this generation to maximize A12 production.

I think it's pretty clear that they've concluded that (at least for now) the tablet market can be served well enough through updates every two years rather than every year. (And they're probably correct; the iPad Pro is still a beast...)

Interesting point(s). It does seem that Apple, at least for now, has very little interest in moving their custom ARM chips into any space other than phones if they have already relegated their tablet products to "also ran" status.

This will probably set a template for the ARM Macs as well. Doing the numbers, people claim Apple Mac volumes are not large enough to support custom SoCs for Macs, but that sort of thinking ignores a variety of ways you can save money. One obvious way is, as I've suggested, to update the SoC every two years. Obviously you'll use the core already designed for iPhone+iPad, though likely with tweaks around the uncore to improve scaling to more CPUs. Finally you use something like EMIB or AMD's interposer, so that you have a single baseline Mac SoC (maybe 4big+4small cores) and you use, I don't know, one in low end laptops, two in pro laptops, and mac minis, three in iMacs, 4 in iMac Pros, 8 in Mac Pros)?

I mean, I guess? Maybe if there's going to be an A12x we'll see how frequently Apple will update their "big" SoCs that could potentially go into future Macs. Right now it looks like they're focusing on phones above all else . . .

Precisely. If i am looking at the available IP from ARM scaling is trivial. DSU supports up to 8 cores (A55, A75, A76) per cluster with 256bit CHI or ACE coherent system ports. Now when licensing the coherent interconnect as well i can connect several DSU clusters coherently and voila - i have a 16, 32 or 64 core ARM system just with licensed IP plugged together. For memory controller i have quite a few options from ARM, Cadence and Synopsis as well.

I would think Cavium and (until recently) Qualcomm used some of that IP to roll out their high core count chips. Not sure how a "glued together" 64-core A12-derived chip would look, but it would be interesting to see. Hopefully they'd abandon their own oddball iteration of big.LITTLE in favor of just the "big" cores though.

 

[frozentundra123456]

AMD achieving parity with Intel in single core would be a monumental achievement....Why wouldn't they after a decade of despair?

Poke pun at it if you find it amusing....Internet cpu forum meltdown....I can't wait to read the silly comments when it happens.

Not "amusing". More like a double standard. I though there were *two* x86 manufacturers that should be trying to compete with Apple/ARM. I dont really consider it a huge achievement to make a big improvement simply because your previous product was so bad.

 

[Kenmitch]

Not "amusing". More like a double standard. I though there were *two* x86 manufacturers that should be trying to compete with Apple/ARM. I dont really consider it a huge achievement to make a big improvement simply because your previous product was so bad.

Compete for what? The Apple ecosystem?

 

[NTMBK]

Compete for what? The Apple ecosystem?

E-peen.

 

[Eug]

For iPhone 10,x (which uses the A11 from last year), the Geekbench 4 scores are around 4300 / 10850 on a good day. Clock speed is 2.4 GHz.

For iPhone 11,x (which is A12 from this year), the best available Geekbench 4 scores are about 4800 single core and about 11350 multi-core.

https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=iphone11

This represents a speed boost of around 5-10% give or take a bit. Clock speed appears to be 2.5 GHz, which is 4.2% higher. So it would appear at least in Geekbench so far, there is not that much of a performance increase, and much of that speed boost is actually due to a clock speed increase.

Note that this is very limited data so take this with a grain of salt. I expect the Geekbench 4 scores to increase with time by as much as about 5% due to expected variation, and if so, that would go along with Apple's claim of about a 15% big core speed boost.

I suspect a good chunk of the increase in transistors (from 4.3 billion transistors to 6.9 billion transistors) may be due to the Neural Engine and the GPU.

 

[name99]

For iPhone 10,x (which uses the A11 from last year), the Geekbench 4 scores are around 4300 / 10850 on a good day. Clock speed is 2.4 GHz.

For iPhone 11,x (which is A12 from this year), the best available Geekbench 4 scores are about 4800 single core and about 11350 multi-core.

https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=iphone11

This represents a speed boost of around 5-10% give or take a bit. Clock speed appears to be 2.5 GHz, which is 4.2% higher. So it would appear at least in Geekbench so far, there is not that much of a performance increase, and much of that speed boost is actually due to a clock speed increase.

Note that this is very limited data so take this with a grain of salt. I expect the Geekbench 4 scores to increase with time by as much as about 5% due to expected variation, and if so, that would go along with Apple's claim of about a 15% big core speed boost.

I suspect a good chunk of the increase in transistors (from 4.3 billion transistors to 6.9 billion transistors) may be due to the Neural Engine and the GPU.

A more informative view is
https://browser.geekbench.com/v4/cpu/compare/7205880?baseline=9816947
which compares against iPhone X.
Are these A12 submissions legit? Who knows?
They seem plausible, in that they reflect a defensible design principle. Single threaded performance is 11% higher, 4% of which is frequency, the rest could just be the larger L1 caches; and the multithreaded is essentially unchanged (a fake would, I guess, have gone all in with boosting that as well). Unchanged multi-threaded may reflect an understanding, after experimenting with both the (very wimpy) Zephyr and the (rather better) Mistral, that there's more value in keeping the small core low power than in boosting its performance.

Memory performance is unchanged, which again seems plausible; just the same LPDDR4 at the same frequency as last year, and a just slightly smarter memory controller? I guess we'll have to wait another year or two before Apple moves to some form of wide low-power DRAM.

The big jumps for Speech Recognition and Dijkstra are also plausible. My analyses of GB overa ll Apple devices show those two are especially sensitive to L2 size (you can test this when you have iPads and iPhones with the same basic core but the iPads have larger L2s).
The other big jumps are AES (that might reflect more instruction fusion; the jump is not large enough to suggest an addition AES engine, but more AES relevant instruction pairs may be fused?)
Also N-body physics, and I've no idea what that's about. Maybe a larger L1D effect, that wasn't visible when I was comparing L2 sizes? (If the dependency chains are long enough, larger L2 won't help but large L1 will.)

 

[ElFenix]

Didn't the Atrix just dock the phone with the laptop? It looks like the Apple patent wants to use the phone as a trackpad. Also, patents don't cover general ideas, but implementations. You can have two patented solutions that accomplish the exact same thing, but if the implementations are different, it's possible for both patents to be valid with neither infringing on each other or preventing someone else from coming up with a third, independent implementation.

there's the whole "obvious" issue. it's a dock. one dock is as obvious as any other.

 

[Thala]

For iPhone 10,x (which uses the A11 from last year), the Geekbench 4 scores are around 4300 / 10850 on a good day. Clock speed is 2.4 GHz.

For iPhone 11,x (which is A12 from this year), the best available Geekbench 4 scores are about 4800 single core and about 11350 multi-core.

https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=iphone11

This represents a speed boost of around 5-10% give or take a bit. Clock speed appears to be 2.5 GHz, which is 4.2% higher. So it would appear at least in Geekbench so far, there is not that much of a performance increase, and much of that speed boost is actually due to a clock speed increase.

Note that this is very limited data so take this with a grain of salt. I expect the Geekbench 4 scores to increase with time by as much as about 5% due to expected variation, and if so, that would go along with Apple's claim of about a 15% big core speed boost.

Sounds about right. 4.2% clock speed boost and roughly 10% IPC boost - which is inline with the claimed 40% efficiency gain. Any larger clock speed boost would diminish the power gain for 7nm.

 

[french toast]

Nope. Kirin 980 are:
- 2x A76 @ 2.8 Ghz
- 2x A76 @ 2.2 Ghz
- 4x A55 @ 1.8 Ghz

Are octacores

Nope. Kirin 980 are:
-2x A76 @ ~2.6 GHz
-2x A76 @ ~2.0 GHz.
-2x A55 @ ~1.8 GHz.

 

[IntelUser2000]

Sounds about right. 4.2% clock speed boost and roughly 10% IPC boost - which is inline with the claimed 40% efficiency gain. Any larger clock speed boost would diminish the power gain for 7nm.

Actually Apple says "Up to 15% faster".

 

[Thala]

Actually Apple says "Up to 15% faster".

And that is precisely what we see here. Both Integer and Floating point scores increase by 15%.

 

[dark zero]

Nope. Kirin 980 are:
-2x A76 @ ~2.6 GHz
-2x A76 @ ~2.0 GHz.
-2x A55 @ ~1.8 GHz.

Indeed, I missed by 200 Mhz. Seems that they downclocked again.

 

[IntelUser2000]

Top score to top score I'm seeing 13% in Int and 15% in FP. In multi-core the difference is 6% for Int and like 2% for FP.

Geekbench is probably the most optimistic as it runs for a few seconds per sub-test.

 

[Eug]

Regarding battery life:

Top score to top score I'm seeing 13% in Int and 15% in FP. In multi-core the difference is 6% for Int and like 2% for FP.

Geekbench is probably the most optimistic as it runs for a few seconds per sub-test.

On the flip side, scores usually improve once people actually have units in-hand. Why? Probably because people turn off everything they can in the OS and then repeatedly run the bench, in order to generate better scores.

The best X scores would probably be representative of that to a certain extent.

I find scores can improve by say 5%.