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

[Eug]

AnandTech's take

In terms of determining the actual process node shrink, the closest valid apples-to-apples comparison we can make are in the small cores and an individual GPU core. Here we see a shrink from 0.53mm² to 0.43mm² in the small CPU cores – representing a 23% reduction. On the GPU core side we see a more significant 37% reduction down from 4.43mm² to 3.23mm².

 

[PeterScott]

AnandTech's take

In terms of determining the actual process node shrink, the closest valid apples-to-apples comparison we can make are in the small cores and an individual GPU core. Here we see a shrink from 0.53mm² to 0.43mm² in the small CPU cores – representing a 23% reduction. On the GPU core side we see a more significant 37% reduction down from 4.43mm² to 3.23mm².

So CPU/GPU shrink, and dies stay about the same size with a massive growth in non defined areas??

 

[Eug]

So CPU/GPU shrink, and dies stay about the same size with a massive growth in non defined areas??

Apple claimed the Neural Engine went from being able to do 0.6 trillion calculations per second (A11) to 5 trillion calculations per second (A12). Faster and more accurate Animojis!!!

 

[Eug]

https://www.phonearena.com/news/App...ks-vs-Galaxy-S9-LG-G7-OnePlus-6-iPad_id109200

 

[Eug]

AnandTech ran SPEC2006 on A12. It does significantly better than older Apple chips and blows everything else on the ARM front out of the water. However, it also seems that where the improvements are the greatest, power utilization spikes significantly.

https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-review-unveiling-the-silicon-secrets/4

The following SPEC figures are declared as estimates, as they were not submitted and officially validated by SPEC. The benchmark libraries were compiled with the following settings:

• 
  
  • Android: Toolchain: NDK r16 LLVM compiler, Flags: -Ofast, -mcpu=cortex-A53
  • iOS: Toolchain: Xcode 10, Flags: -Ofast

On iOS, 429.mcf was a problem case as the kernel memory allocator generally refuses to allocate the single large 1.8GB chunk that the program requires (even on the new 4GB iPhones). I’ve modified the benchmark to use only half the amount of arcs, thus roughly reducing the memory footprint to ~1GB. The reduction in runtime has been measured on several platforms and I’ve applied a similar scaling factor to the iOS score – which I estimate to being +-5% accurate. The remaining workloads were manually verified and validated for correct execution.

The performance measurement was run in a synthetic environment (read: bench fan cooling the phones) where we assured thermals wouldn’t be an issue for the 1-2 hours it takes to complete a full suite run.

SPECint:

SPECfp:

Overall the new A12 Vortex cores and the architectural improvements on the SoC’s memory subsystem give Apple’s new piece of silicon a much higher performance advantage than Apple’s marketing materials promote. The contrast to the best Android SoCs have to offer is extremely stark – both in terms of performance as well as in power efficiency. Apple’s SoCs have better energy efficiency than all recent Android SoCs while having a nearly 2x performance advantage. I wouldn’t be surprised that if we were to normalise for energy used, Apple would have a 3x performance efficiency lead.

This also gives us a great piece of context for Samsung’s M3 core, which was released this year: the argument that higher power consumption brings higher performance only makes sense when the total energy is kept within check. Here the Exynos 9810 uses twice the energy over last year’s A11 – at a 55% performance deficit.

Meanwhile Arm’s Cortex A76 is scheduled to arrive inside the Kirin 980 as part of the Huawei Mate 20 in just a couple of weeks – and I’ll be making sure we’re giving the new flagship a proper examination and placing among current SoCs in our performance and efficiency graph.

What is quite astonishing, is just how close Apple’s A11 and A12 are to current desktop CPUs. I haven’t had the opportunity to run things in a more comparable manner, but taking our server editor, Johan De Gelas’ recent figures from earlier this summer, we see that the A12 outperforms a Skylake CPU. Of course there’s compiler considerations and various frequency concerns to take into account, but still we’re now talking about very small margins until Apple’s mobile SoCs outperform the fastest desktop CPUs in terms of ST performance. It will be interesting to get more accurate figures on this topic later on in the coming months.

 

[John Carmack]

SPEC? It's not noteworthy until I see SYSmark scores.

 

[dark zero]

Wondering if A12X will finally reach Intel Core levels...

 

[Eug]

Wondering if A12X will finally reach Intel Core levels...

What do you mean? Arguably A12 non-X already has

 

[Greyguy1948]

447 DealII is the only fp SPEC benchmark where Apple have got some competition both on power and performance.
Any ideas why?

 

[Greyguy1948]

A link from SPEC:
https://www.spec.org/cpu2006/Docs/447.dealII.html

 

[Greyguy1948]

What do you mean? Arguably A12 non-X already has

If we compare SPECfp2006 the 7 benchmarks for Celeron G530 at 2400 (no turbo) they are:
47 vs 41
16 vs 37
36 vs 37
25 vs 65
36 vs 56
72 vs 101
31 vs 74
so433.milc and 447.deal II are the hardest for Apple.
More computing and less memory & cache reading?

 

[Andrei.]

447 DealII is the only fp SPEC benchmark where Apple have got some competition both on power and performance.
Any ideas why?

It's one of the more execution bottlenecked benchmarks - in its characterization it's most comparable to hmmer. The other cores are relatively competitive here, and the M3 having less FP add latency might be some sort of advantage and why it outperforms it - but I'm just talking out of my ass here based on a first glance.

 

[Greyguy1948]

If we compare SPECfp2006 the 7 benchmarks for Celeron G530 at 2400 (no turbo) they are:
47 vs 41
16 vs 37
36 vs 37
25 vs 65
36 vs 56
72 vs 101
31 vs 74
so433.milc and 447.deal II are the hardest for Apple.
More computing and less memory & cache reading?

And here are the same benchmarks for Atom C2750 at 2400/turbo 2600
milc 20
namd 9
deal II 16
soplex 13
povray 15
lbm 57
sphinx 3 18

 

[Entropyq3]

Minor disclosure: I’m an old computational scientist, and was around at the formation of SPEC.
The original purpose of the SPEC suite was to provide the computational community at the time with something better than vendor FLOPS numbers, in practice necessitating that someone ran your code on the machines of interest which was awfully limiting. It was decided that for data to be put into the database, not only did you have to present all subtest scores, but also what compiler, and compiler settings you used.
Looking back, maybe the biggest overall contribution of the SPEC suite is that it put the spotlight on the importance of compilers, and helped push the quality of generated code across the industry.
Even if that kind of full disclosure may be messy for forum posts, it is always a good idea to put a link to where this information can be found.

Addendum: This is in contrast to just about all benchmarks in PC space which are distributed as binaries, that is, the compiler and settings are the same for all. They still matter of course, and better benchmarks fully disclose what has been used. SPEC is different.

 

[Greyguy1948]

Minor disclosure: I’m an old computational scientist, and was around at the formation of SPEC.
The original purpose of the SPEC suite was to provide the computational community at the time with something better than vendor FLOPS numbers, in practice necessitating that someone ran your code on the machines of interest which was awfully limiting. It was decided that for data to be put into the database, not only did you have to present all subtest scores, but also what compiler, and compiler settings you used.
Looking back, maybe the biggest overall contribution of the SPEC suite is that it put the spotlight on the importance of compilers, and helped push the quality of generated code across the industry.
Even if that kind of full disclosure may be messy for forum posts, it is always a good idea to put a link to where this information can be found.

Addendum: This is in contrast to just about all benchmarks in PC space which are distributed as binaries, that is, the compiler and settings are the same for all. They still matter of course, and better benchmarks fully disclose what has been used. SPEC is different.

Atom CC2750
https://www.spec.org/cpu2006/results/res2014q3/cpu2006-20140617-29940.html

Celeron G530
https://www.spec.org/cpu2006/results/res2011q3/cpu2006-20110908-18550.html

 

[Nothingness]

Addendum: This is in contrast to just about all benchmarks in PC space which are distributed as binaries, that is, the compiler and settings are the same for all. They still matter of course, and better benchmarks fully disclose what has been used. SPEC is different.

The problem is that this also pushed hardware vendors to tweak their compilers for SPEC. Intel icc is notoriously known to be too good to be honest for instance.

The other point of interest in binary only benchmarks is that it also matches what you live as an end user: you have an app that already is compiled, and won't be recompiled with new flags for the new shiny CPU.

That being said, I also favor benchmark which source I can read

 

[Eug]

Sites are reporting A12X endowed iPad Pros will be launching soon. Not a huge surprise though, considering it's been almost 500 days since the last iPad Pro release.

Note though that there is no A11X. Last iPad Pro had the A10X.

 

[Ajay]

Sites are reporting A12X endowed iPad Pros will be launching soon. Not a huge surprise though, considering it's been almost 500 days since the last iPad Pro release.

Note though that there is no A11X. Last iPad Pro had the A10X.

Sensible for Apple to do an every two cycle update for iPads, IMHO.

 

[dark zero]

Sites are reporting A12X endowed iPad Pros will be launching soon. Not a huge surprise though, considering it's been almost 500 days since the last iPad Pro release.

Note though that there is no A11X. Last iPad Pro had the A10X.

I see it going at 2.7 Ghz and the Antutu Score breaking the 500K mark

 

[DrMrLordX]

A12X will be a chip to watch. If Apple has any intention of challenging other markets with their in-house designs, that's where they'll start.

 

[Eug]

Hmmm... Apple is buying part of a chip design company for $600 million. Besides ensuring they get continued supply for more power management chips for a few years, that gets them design engineers plus patents. Design facilities too.

https://www.reuters.com/article/us-...logy-in-600-million-dialog-deal-idUSKCN1ML0IJ

SAN FRANCISCO/FRANKFURT (Reuters) - Apple Inc is buying the power-management technology at the heart of its iPhones in a $600 million deal with Dialog Semiconductor that also secures the German-listed company’s role as a supplier to the U.S. tech giant.

The agreement to acquire patents and people from the Anglo-German chip designer is not only unusual, but also the largest of its kind by Apple, whose last sizeable acquisition was the $350 million purchase of Face ID creator PrimeSense in 2013.

 

[Eug]

A12 Bionic

7 nm, presumably TSMC
6.9 billion transistors
512 GB addressable storage
6-core with 2 big and 4 LITTLE cores (No, Apple didn't use the big.LITTLE terminology)
big cores are 15% faster and 40% more efficient
LITTLE cores are 50% more efficient
All 6 cores can be used simultaneously
5 trillion OPs per second for new Neural Engine (Old one did 0.6 trillion.)
4-core GPU

A12X Bionic

7 nm
>10 billion transistors
1 TB addressable storage
8-core with 4 big and 4 LITTLE cores (No, Apple didn't use the big.LITTLE terminology)
All 8-cores can be used simultaneously
5 trillion OPs per second for new Neural Engine
Single-core 35% faster than old iPad Pros (A10X)
Multi-core 90% faster than old iPad Pros (A10X)
7-core GPU
A12X GPU 2X previous GPU, and 1000X original iPad
Xbox One S class GPU in a 94% smaller volume.

 

[ehume]

A12X Bionic

7 nm
>10 billion transistors
1 TB addressable storage
8-core with 4 big and 4 LITTLE cores (No, Apple didn't use the big.LITTLE terminology)
All 8-cores can be used simultaneously
5 trillion OPs per second for new Neural Engine
Single-core 35% faster than old iPad Pros (A10X)
Multi-core 90% faster than old iPad Pros (A10X)
7-core GPU
A12X GPU 2X previous GPU, and 1000X original iPad
Xbox One S class GPU in a 94% smaller volume.

Is there a news site where you are getting this?

 

[Eug]

Is there a news site where you are getting this?

Apple.com Live Stream

 

[Paratus]

Is there a news site where you are getting this?

ARS has a livestream going on as well