[TechTastic] A10X Geekbench 4 Score

[Arachnotronic]

4236 single core/6588 multicore.

https://www.techtastic.nl/tablets/apple-a10x-soc-voor-nieuwe-ipads-blaast-concurrentie-omver/

This is about what I'd expect in single core from the top bin of Kaby Lake-Y, multi-core still trails a bit (lack of SMT). Impressive stuff.

Intel had better start accelerating the pace of its per-core improvements in ~5W chips or Apple is going to run right past them by the time the A11X rolls around.

EDIT: This source is not trustworthy, see post later in the thread.

 

[jpiniero]

Intel had better start accelerating the pace of its per-core improvements in ~5W chips or Apple is going to run right past them by the time the A11X rolls around.

I'm hopeful Cannon Lake can deliver this; but that's still a year away. The problem with Core M is that it is still drawing too much to really be used in a tablet and throttles.

 

[Arachnotronic]

I'm hopeful Cannon Lake can deliver this; but that's still a year away. The problem with Core M is that it is still drawing too much to really be used in a tablet and throttles.

yep, Core M really needs to be a grounds-up, tailored-for-the-segment SoC. Right now it's Core U with tighter chip packaging.

 

[stingerman]

Hey you know what, Apple's processors are using low voltage designs and showing great performance. It really has you wonder, if they use higher power designs what they can build.

The question: is the A10X a higher clocked A10 and/or on the 10nm node which TSMC already has in production. The iPad Pro would make an interesting candidate for a first to get the 10nm part.

 

[kraatus77]

Apple's cpus are running Taylormade OS specially for them. run them on same OS as other arm/x86 cpus and you will see true difference between them. until then those high GB scores are meaningless and borderline fake madeup/paid by apple to sell their products.

heck you can do some kernel optimizations on android and improve your scores by 5-10% on same cpu. who knows what kind of specific kernel stripping/optimizations are the reason behind high benchmark scores ? and not the cpu architecture itself.

i'll believe apple's performance when they replace intel cpus on their products with their own. until then it's just a marketing gimmick.

 

[StrangerGuy]

Apple's cpus are running Taylormade OS specially for them. run them on same OS as other arm/x86 cpus and you will see true difference between them. until then those high GB scores are meaningless and borderline fake madeup/paid by apple to sell their products.

heck you can do some kernel optimizations on android and improve your scores by 5-10% on same cpu. who knows what kind of specific kernel stripping/optimizations are the reason behind high benchmark scores ? and not the cpu architecture itself.

i'll believe apple's performance when they replace intel cpus on their products with their own. until then it's just a marketing gimmick.

Since we in conspiracy theory territory here, why not just claim all the benchmarks are faked and Apple also paid Intel, QC, Samsung, Mediatek, ARM to shut them up about their supposedly bad chips.

 

[kraatus77]

Since we in conspiracy theory territory here, why not just claim all the benchmarks are faked and Apple also paid Intel, QC, Samsung, Mediatek, ARM to shut them up about their supposedly bad chips.

Or you know, they can just use their cpus on products which are using intel cpus atm. since their cpus are better than intel according to GB? why pay intel for inferior cpus ?

but no00, why do that because it will get beaten once they run a full fledged pc OS. it's pretty simple why they aren't using their own cpus on products which have full OS.

there's no conspiracy theory here, apple's soc isn't running same os as other arm/x86 cpus, thus they perform like that that. it's like pc vs console. console can extract more performance out of same hardware because specific optimizations on software side.

you can't really compare them until both run on same os. simple as that.

 

[asendra]

How reliable is that website? Seems...early? for this kinds of leaks.

Or you know, they can just use their cpus on products which are using intel cpus atm. since their cpus are better than intel according to GB? why pay intel for inferior cpus ?

Software compatibility?

If you don't think you can compare iOs and macOS performance then you don't know much about how they are built. They share a whole lot of the underlying system, and macOS runs on Intel. Also, you can't use the same excuse as Windows that macOS is bloated in order support a huge variety off SKUs...so yeah, you can make educated comparisons.

Its very similar to how Ubuntu has ARM and x86 versions, and Phoronix has been doing performance comparisons of those since ages.

Also, GeekBench 4 was released on September 1, just one week before the iPhone 7 with the A10 and iOs 10 were released, but sure, they are specifically optimised for it.

Last thing, to just put it out there, GeekBench developer was an INTEL!! employee, ffs.

 

[kraatus77]

Heh, i never said anything about GB but the operating system. it can do wonders. all i said is because of different platforms you can't really compare them and determine which cpu is better just because GB score says so.

you can say whatever you want but until apple uses their own cpus on all products. everything is just marketing mumbo jumbo.

Last thing, to just put it out there, GeekBench developer was an INTEL!! employee.

Right, he was. not anymore. not that i have anything against GB for that matter.

for the last time, run a10/x, sd820, core ix whatever on same operating system then run GB on top of it. you will see their true performance on GB (notice i haven't said a10 is slower or faster) . also do run actual real world productivity apps and then compare. you will get a better idea.

 

[MarkizSchnitzel]

Wouldn't a real world test, or, say, video editing of the same material, and some burty tasks be the only thing that matters here, because UX is the only thing that can be directly compared apples to apples?

I mean, why would Intel be in trouble when Apple does not sell these CPUs to OEMs to put into PCs?
Ok, so they might loose some if Apple switches to ARM on MAC, but that's it, no?

 

[Nothingness]

you can't really compare them until both run on same os. simple as that.

You don't know what you're talking about, simple as that. Geekbench has no dependency on the OS, more on the compiler. But we can't stop you believing they are coming to take you

 

[kraatus77]

Selective reading is dangerous you know.

 

[Nothingness]

Selective reading is dangerous you know.

So explain us how the OS will impact the score of a benchmark that is not using anything in the OS.

Do you really think Apple is detecting the benchmark and is pushing clocks higher than for other tasks? If so, it's easy to test: run the benchmark in loop and see when the score degrades, which should happen very quickly due to thermal throttling.

If you have other ideas, enlighten us

 

[kraatus77]

What's the point ? you will again read what you want and will go on with selective reading. but i'll try one more time.

as i already said, you can't compare console to pc just by looking at specifications. because there's alot of optimization goes to extract performance out of it on console. tw3 runs fine on 900p ps4's hd7850 equivalent gpu. but it cannot deliver 100% same performance on on 2ghz ( whatever it is on ps4) bulldozer + hd7850 on pc.

same way you can't compare ios vs android's GB score. android isn't optimized for 1 single soc just like pc. ios and a10/x are made for each other. obviously it will have alot less overhead compared to android running on sd820.

all i'm saying is, without running on same OS, GB score doesn't prove anything. if it was indicative of real performance than apple would have already ditched intel cpus which they haven't yet. and when they do i'll be the first to admit apple's cpus are better than intel.


nothing to do with GB itself or clocks etc. but mostly OS overhead and specific optimizations on governor/scheduler/OS side. which can be significant.

 

[Andrei.]

same way you can't compare ios vs android's GB score. android isn't optimized for 1 single soc just like pc. ios and a10/x are made for each other. obviously it will have alot less overhead compared to android running on sd820.
....
nothing to do with GB itself or clocks etc. but mostly OS overhead and specific optimizations on governor/scheduler/OS side. which can be significant.

You have no idea of what you're talking about.

OS "overhead" is far less than 1%, and then it's just *maybe* DVFS overhead and nothing with schedulers. GeekBench doesn't touch anything on the side of the OS, it's simply native code and instructions which as Nothingness mentioned is affected by nothing else other than compilers. It's an as valid apples to apples comparison as you can have. Are you going to complain about SPEC being irrelevant as well? Because it has the same "OS overhead" problems that you claim GB has. I guess we need bare metal benchmarks to shut up some of those inane arguments.

 

[Qwertilot]

yep, Core M really needs to be a grounds-up, tailored-for-the-segment SoC. Right now it's Core U with tighter chip packaging.

I wonder if they didn't produce it with the plan to have it out on 10nm rather earlier than they've ultimately managed. It'll make much more sense then, although it may take another shrink past that before the design is really in its sweet spot at CoreM sorts of TDP.

 

[kraatus77]

snip

why apple still uses intel cpus. please explain.

why use some other inferior product when you have something better you are already making ?

Funny you left out my console example because it proves my theory.

 

[Arachnotronic]

why apple still uses intel cpus. please explain.

why use some other inferior product when you have something better you are already making ?

Intel CPUs aren't inferior, but with A10X it looks like Apple will roughly match Kaby Lake-Y performance.

 

[Nothingness]

why apple still uses intel cpus. please explain.

why use some other inferior product when you have something better you are already making ?

Because, as was already pointed to you, x86 compatibility matters. And before you can pay the price of runtime translation, you need to be at the very least twice faster, which is not the case (and probably never will be).

Funny you left out my console example because it proves my theory.

Your console example misses the point completely: the middlewares and libraries are not the same on consoles and on PC. GB doesn't depend on such libraries.

 

[kraatus77]

Because, as was already pointed to you, x86 compatibility matters. And before you can pay the price of runtime translation, you need to be at the very least twice faster, which is not the case (and probably never will be).

So apple's cpus aren't better. thanks for proving again.

Your console example misses the point completely: the middlewares and libraries are not the same on consoles and on PC. GB doesn't depend on such libraries.

lol, and it's my time to say "you have no idea what you are talking about" and i'm not even going to explain this anymore because firstly too much selective reading. and second apple is still using intel.

you guys can live in your wonder land where apple cpus are better than intel. unfortunately even apple disagrees

 

[witeken]

4236 single core/6588 multicore.

https://www.techtastic.nl/tablets/apple-a10x-soc-voor-nieuwe-ipads-blaast-concurrentie-omver/

This is about what I'd expect in single core from the top bin of Kaby Lake-Y, multi-core still trails a bit (lack of SMT). Impressive stuff.

Intel had better start accelerating the pace of its per-core improvements in ~5W chips or Apple is going to run right past them by the time the A11X rolls around.

Lol, no, I would rather look at it a different way: Apple is simply catching up to Intel.

You have a bad memory? When Intel launched Core m, it was twice as fast as Apple A8. It took Apple a whopping two (2!) generations to catch up to Broadwell-y's peak (~2.4GHz, so still a little lower than BDW-Y 2.7) performance... But now Intel is already at 3.6Ghz with Kaby Lake Y.

Apple's not going to get past Kaby Lake Y any time soon, with their short cells, and of course neither will Intel, although they will increase sustained performance obviously.

Now that I'm talking about this 3.6Ghz, didn't we at some point with Broadwell-y think that it would go to 4Ghz, I remember something like that .

 

[SAAA]

Oh it's so close to my prediction, also I put the speed range at 2.6-3GHz and this looks to be running at 2.8GHz. Crystal ball doing fine!

Btw about the topic of geekbench4... if we take all the recent scores and leaks at face value, most optimistic case for all of them, then we have:

Zen (assuming it was running 1GHz rather than 1.44, thus possibly "confirming" the 3000 points at 2.9GHz)
~1000pt/GHz

KabyLake (likely the same as Skylake, took the best 7500U score in the database)
~1270pt/GHz

A10 (same as A10X? Again picked the best score of database)
~1500pt/GHz

50% over Zen and 18% over Kaby, see why I'm skeptical about this test? A mobile architecture pulling more than any x86 and twice or more the other arm cores...
I hope someone can pull of some comparison that isn't just one bench and settle the question one day.

 

[Eug]

Lol, no, I would rather look at it a different way: Apple is simply catching up to Intel.

You have a bad memory? When Intel launched Core m, it was twice as fast as Apple A8. It took Apple a whopping two (2!) generations to catch up to Broadwell-y's peak (~2.4GHz, so still a little lower than BDW-Y 2.7) performance... But now Intel is already at 3.6Ghz with Kaby Lake Y.

I agree that Apple is just catching up to Intel now for ultrabook class performance. However, I consider that absolutely remarkable. If you had told me five years ago (when I bought my iPad 2), that Apple's ARM chips would be competing with Intel's laptop chips soon, I would've thought you were an ID10T.

 

[cytg111]

Or you know, they can just use their cpus on products which are using intel cpus atm.

- Following Apples moves for the past few years I fully expect this to be their strategy. We can guesstimate as to why but to me, however unlikely, it seems like they are playing the field in a manner where they will continue to have options later on down the road.
Guess what soc will*not* be fabbed at Intels? .

 

[Andrei.]

50% over Zen and 18% over Kaby, see why I'm skeptical about this test? A mobile architecture pulling more than any x86 and twice or more the other arm cores...

What's so unbelievable about a core that's larger than anything out there? Assuming this estimate is anywhere near reality, then the an A10 core without any L2 is larger than a Zen core + L2. It's almost quadruple the size of a M1 core which is currently the fastest other ARM core, and that's on a less dense process. Remember that while it's faster its per-core power is also higher compared to other ARM cores.

I hope someone can pull of some comparison that isn't just one bench and settle the question one day.

GB4 scales the same as SPEC across several CPUs, so for me that's good enough to give it credibility.