Apple A14 Geekbench 5

[matrix07]

Interesting article.

Fun With Charts: Entirely speculative charts about Apple Silicon

Last week I stared directly into the Apple marketing content and speculated about the information it was trying to impart with a WWDC slide. This week, prompted by reader David Hovis, I’d lik…

sixcolors.com

 

[rkuo]

The minimum memory for any AS powered Macs will be 16GB. Anything less will create a real issue. While posters can go on and on about CISC vs. RISC, how RISC is or isn't inherently "better", whether ARM can be a high performance architecture, etc. ad for forever and a day, the truth is, RISC architectures are less instruction dense than CISC architectures. What that comes down to is that you need more instructions to perform a task than CISC architectures do because the instructions are simpler. That drives the requirement for more RAM on a RISC machine, and why I am saying that 16GB will be the minimum for AS Macs, as it currently is for the DTK Mac Mini.

Think of it another way: Would Apple be sending out a system to developers in which the developers port, cross compile, and test their software (or at least verify their software works under Rosetta 2) with 16GB, and then ship consumer AS Macs with half that amount of RAM? Possible, but at least to me, highly unlikely.

It’s extremely common, even expected, for developer machines to have more memory than the target platform to account for all the debugging overhead.

 

[Pressure]

It’s extremely common, even expected, for developer machines to have more memory than the target platform to account for all the debugging overhead.

You are also limited by the 128-bit memory bus and what LPDDR4X memory Micron makes in this instance.

 

[LiE_]

Max tech did some extrapolation based on past performance gains for a A14X.

 

[Pressure]

Max tech did some extrapolation based on past performance gains for a A14X.

That's quite useless as we have no idea what architectural changes they have made to their processor cores or what TDP they will target.

Why would the iPad Pro suddenly have a 10W higher TDP? It makes no sense.

 

[LiE_]

That's quite useless as we have no idea what architectural changes they have made to their processor cores or what TDP they will target.

Why would the iPad Pro suddenly have a 10W higher TDP? It makes no sense.

It’s from Max Tech’s video where he goes into more detail.

 

[ChromeCloud]

The minimum memory for any AS powered Macs will be 16GB. Anything less will create a real issue. While posters can go on and on about CISC vs. RISC, how RISC is or isn't inherently "better", whether ARM can be a high performance architecture, etc. ad for forever and a day, the truth is, RISC architectures are less instruction dense than CISC architectures. What that comes down to is that you need more instructions to perform a task than CISC architectures do because the instructions are simpler. That drives the requirement for more RAM on a RISC machine, and why I am saying that 16GB will be the minimum for AS Macs, as it currently is for the DTK Mac Mini.

Think of it another way: Would Apple be sending out a system to developers in which the developers port, cross compile, and test their software (or at least verify their software works under Rosetta 2) with 16GB, and then ship consumer AS Macs with half that amount of RAM? Possible, but at least to me, highly unlikely.

The assumption that Apple Silicon Macs will need more memory because of the RISC instruction set is wrong. If anything, the difference will be marginal.

99% of the RAM in a computer is used for graphic assets and data structures.

The actual executable code is so small it doesn't even matter.

Just for reference, the actual executable code of Safari on Catalina is 42 KB (yes, Kilobytes).

 

[souko]

It does not make sense that 15W TDP in iPad Pro is passively cooled. According to https://www.techpowerup.com/249160/apples-a12x-shows-us-how-the-arm-macbook-is-closer-than-ever TDP of A12X is 7W...

Another note: iPad Pro comes with 18W charger. This charger is able to charge iPad under load. Which will not be truth if only CPU consumes 15W.

according to: https://www.notebookcheck.net/Apple-iPad-Pro-11-2018-WiFi-64-GB-Tablet-Review.357931.0.html Whole iPad Pro 2018 consumes 15W under max load and max brightness.

 

[vigilant]

The minimum memory for any AS powered Macs will be 16GB. Anything less will create a real issue. While posters can go on and on about CISC vs. RISC, how RISC is or isn't inherently "better", whether ARM can be a high performance architecture, etc. ad for forever and a day, the truth is, RISC architectures are less instruction dense than CISC architectures. What that comes down to is that you need more instructions to perform a task than CISC architectures do because the instructions are simpler. That drives the requirement for more RAM on a RISC machine, and why I am saying that 16GB will be the minimum for AS Macs, as it currently is for the DTK Mac Mini.

Think of it another way: Would Apple be sending out a system to developers in which the developers port, cross compile, and test their software (or at least verify their software works under Rosetta 2) with 16GB, and then ship consumer AS Macs with half that amount of RAM? Possible, but at least to me, highly unlikely.

This just hurt my head. I don’t believe anyone has ever made this conclusion before.

I’d encourage you to go read on the differences between RISC and CISC.

I can assure you that when the Mac was on PowerPC and migrated to Intel, Apple didn’t get a magic boost in memory.

 

[Realityck]

Ok you see tests for single core, multi core, but where is the pairing of a external GPU with ARM? See Metal Benchmarks

per slashgear article about Apple Silicon

Benchmarks revealed how the Apple A12Z inside Apple’s Developer Transition Kit (DTK) is almost on par with integrated GPUs on Intel’s and AMD’s latest-gen processors. And this was still with the benchmark tool running via the Rosetta compatibility layer, which naturally incurs some performance hit. Apple’s first ARM-based Macs are also expected to run on something more powerful than the Apple A12Z and have a more powerful GPU as well.

 

[Waragainstsleep]

It’s from Max Tech’s video where he goes into more detail.

Max tech does not appear to understand much about CPUs. He extrapolates performance based on previous progress and comparisons which is flawed enough but far more sensible than assuming the TDP of a CPU will go up in the same way. The iPad enclosure has the capacity to cool a chip with a given TDP. While Apple can tweak the case and/or the cooling, unless they make it thicker (does that sound likely?), or install a fan (which would kill the battery life), the TDP is going to stay within that limit. There ought to be some headroom to be safe, probably 20W or something like that but they will never go over 16W or so if that's the case.
And jacking up the TDP of CPU has a knock on effect for the charger as mentioned and also the battery. CPU and screen will be the big power consumers in an iPad, jack one of those 50%, expect to lose 25% of your battery life, maybe more.

 

[D.T.]

Grain of salt / normal caveats in place with information like this, that 8259 compute score is pretty stellar given this isn't the X or Z flavor (haven't we historically seen those usually double that performance?)

 

[Kostask]

This just hurt my head. I don’t believe anyone has ever made this conclusion before.

I’d encourage you to go read on the differences between RISC and CISC.

I can assure you that when the Mac was on PowerPC and migrated to Intel, Apple didn’t get a magic boost in memory.

I don't know if I explain it, it would do a better job than this:

RISC and CISC Architecture - Working & Their Applications

This Article Discusses an Overview of What are RISC and CISC Architecture - Working, Differences, Advantages, Disadvantages & Its Applications

www.watelectronics.com

To quote directly from the article (it is a direct copy and paste from that page):

"In RISC, more RAM is required to store assembly level instructions."

When I wrote what I did, what I meant was, in order to accommplish the same task, RISC machines normally require more instructions than CISC machines do (which is what I called instruction density). The article used the Multiply instruction as an example. But even something as simple as an add and store could have been used. In CISC CPU, add and store is often one instruction. In a RISC CPU, it is two. Therefore, you need 2 memory locations for the RISC instructions, and one memory location for the CISC instruction.

I understand that compilers and hand optimization can mitigate this to some extent, but in general, RISC needs more memory than CISC does, and if I can use my own term, the CISC machine has higher "instruction density" than RISC does.*

I also understand that the various accelerator modules that Apple employs can impact that, but those accelerator modules do not come into play with most regular applications or the OS itself.

*=This was known as far back as the original IBM RISC machine, mid to late 1980s.
[automerge]1596146191[/automerge]

It’s extremely common, even expected, for developer machines to have more memory than the target platform to account for all the debugging overhead.

Please let me know at what point Apple is calling the DTK Mac Mini machine as a developer platform. Because I certainly have never seen, read or heard it being presented that way. Apple is saying that this is a "transition kit", and is specifically calling it a "transition kit" over and over. It is designed to allow developers to run apps compiled for AS to run on AS hardware.

As you said, this is the "target platform", not the development platform.

Can you honestly see somebody who is developing on 32+GB High End iMacs, Mac Minis and MacBook Pros just abandoning those platforms to run on a potentially unstable AS Mini? I can't. I can see them running the resultant code on the AS Mini, as it is the only game in town, but there is no reason to move from an established platform that requires just the flip of a software switch to produce AS code.

 

[leman]

When I wrote what I did, what I meant was, in order to accommplish the same task, RISC machines normally require more instructions than CISC machines do (which is what I called instruction density). The article used the Multiply instruction as an example. But even something as simple as an add and store could have been used. In CISC CPU, add and store is often one instruction. In a RISC CPU, it is two. Therefore, you need 2 memory locations for the RISC instructions, and one memory location for the CISC instruction.

These are purely theoretical considerations, not to mention that they are flawed. Instruction density is a complex topic that depends on a variety of factors. You can’t just make a blanket statement like that just because one ISA is load/store while the other one is not. Instruction size, number of registers, ISA ABI, finally the instructions themselves - all this plays a role.

There are peer-reviewed papers comparing ARM64 and x86-64 code density, using real world code. Look them up.

P.S. The link you give is outdated by 20+ years. Long time ago, RISC and CISC were different hardware designs (as that link describes). That distinction has completely lost its meaning with modern High-performance CPUs. All of them are microcode controlled super-scalar chips magnitudes of complexity higher than the diagrams in your link suggest.

 

[Krevnik]

I’m not certain that’s a forgone conclusion. The March iPad Pro has lower clock speeds than the iPhone 11 I believe. More cores equals more heat than can’t be dissipated?

The iPhone and iPad Pro also use a passively cooled chip with zero air flow, limiting heat dissipation.

There's an awful lot of headroom to play with in the current Mac hardware. Higher clock speeds, more cores, whatever Apple wants to do to try to improve performance, they have options that they don't have on the A-series SoCs. They could probably go fanless on the low end systems, but it'd be tough to keep up with the current i9 16" MBP without some active cooling.

As you said, this is the "target platform", not the development platform.

Can you honestly see somebody who is developing on 32+GB High End iMacs, Mac Minis and MacBook Pros just abandoning those platforms to run on a potentially unstable AS Mini? I can't. I can see them running the resultant code on the AS Mini, as it is the only game in town, but there is no reason to move from an established platform that requires just the flip of a software switch to produce AS code.

Sometimes, it's easier to just do the compile on the system itself and debug from there. Building and remote debugging hasn't been the ideal way to go in my experience. Not to mention trying to shuttle builds around is just annoying.

If my incremental builds are quick enough, then I'd just use a special branch for the Apple Silicon changes on the DTK. If not, I'd probably group up changes in that special branch from my more powerful machine, and then do the build locally on the DTK when I needed to spend time on the hardware itself validating things.

That said, memory usage has been dominated by things other than code for a long time. And the memory manager has tricks to avoid keeping the whole set of code an app bundle might contain in memory all at once. Larger apps that use shared libraries like Qt can leverage this to have a smaller code memory footprint than they have on disk. Because of that, the small increase in code size may not even have a corresponding impact on RAM usage of the code, since code is still swapped in using 4KB page sizes. Let alone having enough of a difference in total RAM usage by the app to be noticed compared to the considerable RAM resources being used to hold data.

 

[UltimateSyn]

I really hope they make a 15-16” MBA, or make the 16” MBP considerably lighter using these PPW gains.

 

[Danny82]

Bored because seems like all discussion has been done about up coming Apple silicon and no one is hype anymore.. what a long wait

Just sharing further rumors to spark the excitement~

 

[Six Mac Abs]

Bored because seems like all discussion has been done about up coming Apple silicon and no one is hype anymore.. what a long wait

I feel the same. I've been trying to hunt down more info on the net since info dried up here. Genuinely interested and excited to see what A14 and Apple Silicon Macs will bring. Thanks for sharing what you found ??.

Do people think the MacBooks will include a variant of the A14 chip, or do you think they will release a new line specifically for desktops / laptops?

 

[leman]

Do people think the MacBooks will include a variant of the A14 chip, or do you think they will release a new line specifically for desktops / laptops?

I think it will be a variant based on A14 architecture, but with more cores, cache, memory controllers and additional dedicated hardware such as IO controllers.

 

[Cookie18]

I feel the same. I've been trying to hunt down more info on the net since info dried up here. Genuinely interested and excited to see what A14 and Apple Silicon Macs will bring. Thanks for sharing what you found ??.

Do people think the MacBooks will include a variant of the A14 chip, or do you think they will release a new line specifically for desktops / laptops?

Rumour has it that the first ones will include an A14X chip.

 

[Danny82]

I feel the same. I've been trying to hunt down more info on the net since info dried up here. Genuinely interested and excited to see what A14 and Apple Silicon Macs will bring. Thanks for sharing what you found ??.

Do people think the MacBooks will include a variant of the A14 chip, or do you think they will release a new line specifically for desktops / laptops?

Yeah, rumors leading to the mac apple silicon to be 8p core and 4e core.. i know it does not work that way but since no one can know anything till they release.. i just want to hype up myself for fun.. so would always take leaked single core geekbench perform x8.. and always amaze with the power it will bring if extrapolate.. ?

 

[JohnnyGo]

Rumour has it that the first ones will include an A14X chip.

IMHO Apple will limit the A14X for the iPad Pro and have a new line of SOCs for the MacBook/MBPs.

This new SOC line will leverage the same technology in the A14 but with many more cores, higher frequencies, higher TDPs as MacBooks have much larger batteries (and more cores will need more power), enhanced memory controllers as the base version should be 16Gb but they will need 32 and 64Gb options.

It’s both more units (core count) AND smarter/better units (different/additional features) inside this new SOC

 

[Woochoo]

This smells fishy.

It may be fake, but results are realistic. A13 already improved between 15% and 25% over A12 in single core, and this will be a new arch on 5nm, clocking at 3.0GHz as it shoes seems totally feasible. Apple leaps have been huge, they are basically wiping all ARM competition and keep in mind those SoCs have way more die size than AMD or Intel CPUs.

 

[ccollinsradio]

Max tech does not appear to understand much about CPUs. He extrapolates performance based on previous progress and comparisons which is flawed enough but far more sensible than assuming the TDP of a CPU will go up in the same way. The iPad enclosure has the capacity to cool a chip with a given TDP. While Apple can tweak the case and/or the cooling, unless they make it thicker (does that sound likely?), or install a fan (which would kill the battery life), the TDP is going to stay within that limit. There ought to be some headroom to be safe, probably 20W or something like that but they will never go over 16W or so if that's the case.
And jacking up the TDP of CPU has a knock on effect for the charger as mentioned and also the battery. CPU and screen will be the big power consumers in an iPad, jack one of those 50%, expect to lose 25% of your battery life, maybe more.

For real. His YouTube is big into 'niching', so when he gets one video that performs well (ie, an Apple Silicon one), he makes 6 or 7 other videos guessing about other things, because the topic is hot. By the fourth video, nearly 80% of the content is strictly from the older videos.

Most channels that behave this way shorten their life span.

Also, the AppleInsider graphic looks all kinds of sketchy with the superblur on the right side. Also, for the gains shown, I'd say that isn't a lot considering the 600 Mhz clock advantage on the blurry photo.

 

[the8thark]

Also, the AppleInsider graphic looks all kinds of sketchy with the superblur on the right side.

All these images are just that, images. Zero source links shown so we can varify the image and read up more info on the image if we want to.