First A14 Geekbench Result from Ice Universe on Twitter

[thingstoponder]

https://twitter.com/i/web/status/1312230485343514626

Performance looks great. The a12 + 40% claim was correct for iPads, but underplays it in multi core for iPhone. The iPad Air a12 scores higher than the iPhone version so this score is more like 70% faster than the iPhone XS and 80% faster than the XR, at least on geekbench. Single core is the same for all models and the 40 percent claim is valid there too.

Also first evidence of the iPad Air 4 RAM amount, it’s 4GB.

And people thought this was a minor upgrade...

 

[Birkan]

Here are the actual CPU and Compute benchmark links for A14 for further discussion.

Compared to A13, single-core score is improved around 18% (1336 to 1583), multi-core score is improved around 18% (3539 to 4198), and compute score is improved around 65% (7608 to 12571).

Compared to A12Z, single-core score is improved around 40% (1123 to 1583), multi-core score is slower around 11% (4699 to 4198), and compute score is basically same (12479 to 12571).

CPU increases against A13 that are around 18% seems decent but what's more interesting is that they are basically matching 8 core GPU from A12Z with 4 core GPU found in A14. It looks great for the upcoming Apple Silicone based Macs!

 

[EntropyQ3]

Yeah, I wonder a bit at that compute score. It used to be a straight GPGPU test - is the ML block involved now?

 

[Jouls]

Does anybody know why the scores don’t turn up in the iOS results list?

 

[Falhófnir]

Does anybody know why the scores don’t turn up in the iOS results list?

If you mean this page:

iOS Benchmarks - Geekbench

It's an average and needs at least 5 runs submitted before it will show up there. I assume there's some verification needed as well, so as the iPad Air isn't officially available yet they probably won't allow the aggregate score to show up until it is.

 

[leman]

The massive increase in compute performance (if true) could suggest that Apple has reorganized their shader units. Maybe taking some ideas from Imagination A series abs gone super-wide?

 

[Jouls]

It's an average and needs at least 5 runs submitted before it will show up there.

I see. Thanks.

 

[thingstoponder]

Here’s the actual link to the entry. No Metal test yet, very curious for that.

iPad13,2 - Geekbench

Benchmark results for an iPad13,2 with an ARM processor.

browser.geekbench.com

My quick calculation for how this a14 would run on a Mac using Rosetta is 1199 for single core if we apply the same penalty from the native DTK score vs the emulated x86 one. Faster than a12Z running Arm code. I didn’t calculate multi core because the a12Z is 4 big cores in Rosetta and the a14 is 2+4 native.

 

[EugW]

Here’s the actual link to the entry. No Metal test yet, very curious for that.

iPad13,2 - Geekbench

Benchmark results for an iPad13,2 with an ARM processor.

browser.geekbench.com

My quick calculation for how this a14 would run on a Mac using Rosetta is 1199 for single core if we apply the same penalty from the native DTK score vs the emulated x86 one. Faster than a12Z running Arm code. I didn’t calculate multi core because the a12Z is 4 big cores in Rosetta and the a14 is 2+4 native.

The Metal benchmark was posted at Geekbench.com at about the same time as the CPU bench yesterday. I posted it here:

A14 benchmarks

Yes, very fast, but it’s pretty much as expected given what Apple said at the September event. However, GPU is very fast, slightly faster than A12Z, assuming this is truly legit. https://browser.geekbench.com/v5/compute/1581541 This bodes well for the Arm Macs.

forums.macrumors.com

However, here is the direct link:

iPad13,2 - Geekbench

Benchmark results for an iPad13,2 with an ARM processor.

browser.geekbench.com

 

[thingstoponder]

Thanks. It seems to be leveraging the new neural engine or something. Apple just mentioned GPU performance at the keynote so I’m guessing they meant graphics instead of raw compute power.

 

[EntropyQ3]

I don't think the Geekbench 5 compute test leverages the neural engine. I would love for John Poole to verify that though.

 

[Birkan]

A14 as well as A13 does support Apple's machine learning accelerators (AMX) however I'm not sure if they can be leveraged in Metal compute operations or Neural Engine in any case.

One thing I noticed was increased Metal compute scores (going from average of 4402 to 5455) on my iPhone XS on iOS 14 vs iOS 13. But since my iPhone XS only has A12 which doesn't support AMX, I believe this change is mostly thanks to optimised GPU drivers/pipeline. That's why I picked devices running iOS 14.0.1 for both A12Z and A13 comparison reference points.

Here are the actual CPU and Compute benchmark links for A14 for further discussion.

Compared to A13, single-core score is improved around 18% (1336 to 1583), multi-core score is improved around 18% (3539 to 4198), and compute score is improved around 65% (7608 to 12571).

Compared to A12Z, single-core score is improved around 40% (1123 to 1583), multi-core score is slower around 11% (4699 to 4198), and compute score is basically same (12479 to 12571).

CPU increases against A13 that are around 18% seems decent but what's more interesting is that they are basically matching 8 core GPU from A12Z with 4 core GPU found in A14. It looks great for the upcoming Apple Silicone based Macs!

 

[ArPe]

. It looks great for the upcoming Apple Silicone based Macs!

Only if the macOS install/build and third party apps on the ARM machine has the same optimisation and fluidity as iOS/iPadOS. If it is as cranky and buggy as it has been on Intel then the only initial benefits are laptops with long battery and low temperature.

We would also need 4 to 8 big cores match Intel multiprocessor performance. But we all know about what I’ve said in this post.

I think the Apple Silicon laptops will sell very well because most users just want something that has long battery and no noticeable heat. Those aren’t power users and don’t know anything about things like benchmarks.

 

[fokmik]

These numbers will increase slightly, since the ipad air has more pixels to push
So the iphone 12 will have better numbers...but

This is strong....this is the base A14 chip, so this is just for the iphones
The ipad pro will have A14Z that will be even better by probably around 20%
And for the first macs even better than the ipad pro
So we can expect that the macbook pro 14" chip will be around 30-50% better than this based a14 chip
Besides that, think about no more heavy heating, think about 14-16 battery life

Magnificent

 

[thingstoponder]

These numbers will increase slightly, since the ipad air has more pixels to push
So the iphone 12 will have better numbers...but

This is strong....this is the base A14 chip, so this is just for the iphones
The ipad pro will have A14Z that will be even better by probably around 20%
And for the first macs even better than the ipad pro
So we can expect that the macbook pro 14" chip will be around 30-50% better than this based a14 chip
Besides that, think about no more heavy heating, think about 14-16 battery life

Magnificent

You’re underplaying it. 2 more performance cores of the a12X/Z will have much more than 20 percent performance in multi core. And if the Mac chip has 8 big cores like Bloomberg said then it will be way more than 50% over the a14.

A12Z in iPad Pro is already 60% more than the a12 in the Air 3.

 

[Birkan]

Only if the macOS install/build and third party apps on the ARM machine has the same optimisation and fluidity as iOS/iPadOS. If it is as cranky and buggy as it has been on Intel then the only initial benefits are laptops with long battery and low temperature.

We would also need 4 to 8 big cores match Intel multiprocessor performance. But we all know about what I’ve said in this post.

I think the Apple Silicon laptops will sell very well because most users just want something that has long battery and no noticeable heat. Those aren’t power users and don’t know anything about things like benchmarks.

Actually, most of the apps should run out of the box under Apple Silicon as Apple made some default optimisations for apps built with macOS 10.15 and earlier SDKs. They are automatically enrolled to some compatibility conditions. If they are built with macOS 11.0 SDK, those compatibility checks shouldn't be there and it should offer even better performance. Of course, these only apply to actual Metal based applications. OpenGL/CL still seems to be supported even though they are deprecated. That's where most users might see some problems.

That's true, we will definitely need more than 4 GPU cores for Apple Silicon Macs. I was just doing an Apples to Apples comparison between A13/12Z to 14. I didn't want to speculate anything regarding number of CPU, GPU, NPU cores, their frequencies, total TDP of the devices as they will depend on what Apple chooses for each category, i.e. MBA, 13 inch MBP, 16 inch MBP etc. But IMHO, rumoured 8 High Performance core, 4 Efficiency core SOC with 8-16 GPU cores should be pretty decent upgrade for a 13 inch MBP class device.

 

[fokmik]

You’re underplaying it. 2 more performance cores of the a12X/Z will have much more than 20 percent performance in multi core. And if the Mac chip has 8 big cores like Bloomberg said then it will be way more than 50% over the a14.

A12Z in iPad Pro is already 60% more than the a12 in the Air 3.

again, im playing this game safe...what can we get at the minimum...we dont know what and how big the diff will be...for now are just rumours. If you make your hopes too hight, your disappointment will be bigger
but again, remember, those big mac chips will not come into the macbook/macbook air family....those will come next year into the 16" mbp, bigger imac
So, dont expect the next macbook air, or the base imac to have more than 50% than the a14 like i said
but again, i hope i will be wrong...but again, i dont see Apple put money into an specific, different chip just for macbook air, i see the macbook air to have similar A14/A14X chip....since even with that, will crush the current Intel macbook air, from cpu raw power, to the igpu raw power, heat, and battery life

Nevertheless, just 1 more month until Apple could announce the new macs based on the 5nm arm architecture

First year after so many, that mac is on top

 

[ChrisA]

Yeah, I wonder a bit at that compute score. It used to be a straight GPGPU test - is the ML block involved now?

I think this comment went right over some read's heads. If there is an ML block or any other non-ARM block invloved then the high score says rather nothing about the ARM CPU core.

Maybe it requires some explaination... If you are writing software and have to do something like multiply and add a few million numbers you can do this two or three ways. One is to use code on the CPU to fetch, multiple, add and store the numbers using CPU registers and the multiply and add units inside the CPU. The second way is to use the CPU to fetch the numbers and then pass them off to some extrnal hardware that is not inside the ARM core to do the actual math. A third way to to tell this external-to-ARM hardware where the numbers are and what to do with them.

If the third method was used, we don't learn about the ARM core but we do learn about Apple Silicon.

I think many consumers don't understand the difference between "Apple Silicon" and "ARM". ARM is just the CPU. Apple Silicon includes the CPU and also that external hardware that can be told where the numbers live and what needs to be done with them. The benchmark is for Apple Silicone.

This is what "is the ML block involved.." question as asking.

Is this Geek Bench software open sourse? If so it would be easy enough just to read it and see what it is actualy measuring.

 

[fokmik]

On the A12Z i managed to test and run on the latest beta...a game called i think league of heroes, at 60 fps, on 1080p
On beta 3, never could run it with more than 46 fps
So, Rosetta 2 is improving, nothing is final

 

[fokmik]

Start hoping by this Monday...i guess this coming week, we will have the release of the ipad air 4 and the announcement of the next digital event for 13th October

 

[ChrisA]

On the A12Z i managed to test and run on the latest beta...a game called i think league of heroes, at 60 fps, on 1080p
On beta 3, never could run it with more than 46 fps
So, Rosetta 2 is improving, nothing is final

Games might be a poor test of the ARM core as they tend to be GPU dependent.

A better test of the ARM core would be to build a large software project. Try compiling the PostgreSQL DBMS or some other system with more than a 100K LOC. On Linux systems a kernal build is a good hardware benchmark but that would not work on xcode. Maybe have gcc do a full native build on Intel mac then on AS.

 

[fokmik]

Games might be a poor test of the ARM core as they tend to be GPU dependent.

A better test of the ARM core would be to build a large software project. Try compiling the PostgreSQL DBMS or some other system with more than a 100K LOC. On Linux systems a kernal build is a good hardware benchmark but that would not work on xcode. Maybe have gcc do a full native build on Intel mac then on AS.

again, ive tested that because the gpu is the big question for me....what Apple can do, and will do , to replace the amd dGpu from the 16" Mbp, imacs, and mac pro because there are the big differences ....so an iGpu from 2 years, under emulation is already good enough, and again i was doing that to see if beta after beta Rosetta 2 is still improving
Based on what ive seen, Apple is doing very well on cpu side, no questions there (only for the Mac Pro, but that will come in 2022 probably)

 

[leman]

Games might be a poor test of the ARM core as they tend to be GPU dependent.

A better test of the ARM core would be to build a large software project. Try compiling the PostgreSQL DBMS or some other system with more than a 100K LOC. On Linux systems a kernal build is a good hardware benchmark but that would not work on xcode. Maybe have gcc do a full native build on Intel mac then on AS.

A12 and A13 have been exhaustively tested by Anandtech using industry standard benchmark tools. I don't think that compiler performance is the best CPU test around, since it heavily depends on storage and RAM. That said, Apple CPUs do extremely well in compiler-based tests in both Geekbench and SPEC. Massive caches and excellent branch prediction work wonders here, and 16Kb memory pages don't hurt either.

 

[jeanlain]

I think this comment went right over some read's heads. If there is an ML block or any other non-ARM block invloved then the high score says rather nothing about the ARM CPU core.

AFAIK, the ML core can only be accessed via coreML (and possibly other high-level APIs), which is unlikely to be used by the CPU tests.
For the Metal compute test though... I suppose this tests uses compute shaders, which are processed by the GPU. But I wonder if the neural engine could be used behind the scene?

 

[jeanlain]

a game called i think league of heroes

This doesn't appear to be a macOS game. This runs on iOS.