Apple A9X the new mobile SoC king

[IntelUser2000]

BTW, 95% people don't know the history of DEC P.A. branch.

That's why they don't understand that P.A.(accquired by Apple) is the most dangerous opponent to Intel.

Now I don't know much about P.A, but the worst thing to assume is that a limit of one(in this case Intel's ability to make faster CPUs) is the limit of another.

Intel spent the past 5 years frantically trying to get into mobile(and affecting their PC line) in vain. And losing billions in the process, though not significant of an impact at this time.

Intel has kept hold of the x86 market through bullying and unfair practices. The one that "motivated" Intel to do better used to be AMD. That is a failure. Entrance by others like Nvidia is shut down by legal battles. Now maybe people will get to see their grip on the x86 market broken by a force that they can't stop - mobile.

 

[Rakehellion]

Relatively unknown benchmark comes to prominence by being the poster boy for the 64-bit performance boost of the A7 cpu and despite clear problems with it in 2013, it's still somehow the primary benchmark used today.


Of course, the only problem with that is despite Anand's insistence, how would you even know if iPhone does the same thing given Apples control of the entire system? There are many ways to recognize an application other than using its name. Or with Geekbench, where the "cheating" comes at the code and compiler level.

I think the real problem is that you believe this even though you have zero evidence to support your claim. You made up a lie that you believe.

 

[catinthefurnace]

Here is another example of why Geekbench is terrible and favours Apple designs rather heavily. (Apple improves memory bandwidth drastically between generations.)

50% performance boost with 50% faster memory. Pretty much scales linear with memory speed.

While a more computational heavy physics bench shows much less.

This is also reflected with the iPhone 6 Plus. As soon as it comes to real computational work it tanks massively.

I'll choose to be friendly and give you the benefit of the doubt. You must have made a mistake when complaining that Geekbench's MEMORY SCORE scales linearly with memory bandwidth.

Please show how Geekbench's OVERALL SCORE scales with memory bandwidth increases. Since that is obviously the only thing even remotely relevant to your proposed complaint.

 

[Thala]

50% performance boost with 50% faster memory. Pretty much scales linear with memory speed.

Lol. What precisely do you expect from a memory benchmark? I guess you completely missed the fact, that the linked picture just shows the memory sub-score of Geekbench...
So actually your argument goes in favor of Geekbench then...

 

[Accord99]

I think the real problem is that you believe this even though you have zero evidence to support your claim. You made up a lie that you believe.

No, the real problem is that Apple fans are using a set of benchmarks that would have been laughed out of the room in 1999 to make assertions that favor their own point of view.

 

[Phynaz]

Relatively unknown benchmark comes to prominence by being the poster boy for the 64-bit performance boost of the A7 cpu and despite clear problems with it in 2013, it's still somehow the primary benchmark used today.


Of course, the only problem with that is despite Anand's insistence, how would you even know if iPhone does the same thing given Apples control of the entire system? There are many ways to recognize an application other than using its name. Or with Geekbench, where the "cheating" comes at the code and compiler level.

You realize you can compile and deploy your own apps to iPhone, right?

 

[Accord99]

You realize you can compile and deploy your own apps to iPhone, right?

Yes, and despite that we almost never see reviews test out other benchmarks or their own custom workloads on mobile devices other than the occassional spec run; unlike during the periods of AMD-Intel CPU competition.

 

[catinthefurnace]

No, the real problem is that Apple fans are using a set of benchmarks that would have been laughed out of the room in 1999 to make assertions that favor their own point of view.

Similar in fact to the way you're being laughed out of this room right now for not providing anything specific to support your claims.

 

[scannall]

No, the real problem is that Apple fans are using a set of benchmarks that would have been laughed out of the room in 1999 to make assertions that favor their own point of view.

Oh good grief. If you don't like any of the many benchmarks, then take off the tinfoil hat and write an open source one yourself. Let the world stand in awe at your brilliance.

 

[gdansk]

If anyone owns SPEC CPU2006, post results and this thread can be dead, either way

 

[Space69]

It's not possible to conduct cross platform performance comparisons with Geekbench as has been said multiple times. The data set is different - usually 4 times larger on desktop compared to mobile. Different algorithms between desktop and mobile versions. It's very depended on compilers - here's the differences between Xcode 5 (on the left) version of Geekbench and Xcode 7 (right) of Geekbench: http://browser.primatelabs.com/geekbench3/compare/3571781?baseline=3572808

 

[ShintaiDK]

It's not possible to conduct cross platform performance comparisons with Geekbench as has been said multiple times. The data set is different - usually 4 times larger on desktop compared to mobile. Different algorithms between desktop and mobile versions. It's very depended on compilers - here's the differences between Xcode 5 (on the left) version of Geekbench and Xcode 7 (right) of Geekbench: http://browser.primatelabs.com/geekbench3/compare/3571781?baseline=3572808

:thumbsup:

 

[Thala]

The data set is different - usually 4 times larger on desktop compared to mobile. Different algorithms between desktop and mobile versions.

Anything to back up this claim?

It's very depended on compilers

That's a property of every code. So nothing to see here. Practically you would hope, that a benchmark would always use the latest compiler.
This holds in particular for ARM, as the AArch64 back-ends are not as mature as the x86-64 back-ends.

 

[thunng8]

It's not possible to conduct cross platform performance comparisons with Geekbench as has been said multiple times. The data set is different - usually 4 times larger on desktop compared to mobile. Different algorithms between desktop and mobile versions. It's very depended on compilers - here's the differences between Xcode 5 (on the left) version of Geekbench and Xcode 7 (right) of Geekbench: http://browser.primatelabs.com/geekbench3/compare/3571781?baseline=3572808

Difference in dataset size affect run times, not results. Primate said in their tests there was less than 1% variance. Nevertheless, geekbench 4 will have same dataset sizes to because mobile is now very fast so there is no need to have smaller datasets (the reason why is was smaller was because of runtime considerations) Don't know hat you mean by algorithms. The tests all test the same things, but some do test inbuilt accelerator in the cpu.

And what test is not affected by what compiler is used? Better, faster compiler will run tests faster. It is a no brainer.

Btw spec2006 is the test that has huge compiler dependency. Difference between gcc and icc on Intel hardware could be up To 50% variance.

 

[Space69]

Anything to back up this claim?

http://www.primatelabs.com/geekbench3/workloads.pdf

Difference in dataset size affect run times, not results. Primate said in their tests there was less than 1% variance.

How convenient. If the variance is less than 1%, then why have differences? What is the impact on throttling?

And what test is not affected by what compiler is used? Better, faster compiler will run tests faster. It is a no brainer.

Btw spec2006 is the test that has huge compiler dependency. Difference between gcc and icc on Intel hardware could be up To 50% variance.

Well that sounds almost like the 70% variance when using Xcode. Highly optimized asm is a very good start, since we're interested in architecture comparisons and NOT compiler performance. Why introduce yet another unknown by using updated compilers in the dataset? You can't use the same algorithm if you compare fix functions vs sw.

 

[jfpoole]

How convenient. If the variance is less than 1%, then why have differences? What is the impact on throttling?

When Geekbench 3 was released, there was a huge delta between desktop and mobile performance. Running the desktop dataset on mobile devices would increase runtimes to a point where our users would become impatient. Running the mobile dataset on desktops would decrease runtimes to a point where timer precision could be an issue on some platforms.

Now that the delta between desktop and mobile is much smaller we'll use the same dataset on desktop and mobile for Geekbench 4.

Also, all workloads use the same algorithm implementation regardless of platform. e.g., the Sobel workload uses the same source code for desktop and mobile -- only the input image varies.

Well that sounds almost like the 70% variance when using Xcode. Highly optimized asm is a very good start, since we're interested in architecture comparisons and NOT compiler performance. Why introduce yet another unknown by using updated compilers in the dataset? You can't use the same algorithm if you compare fix functions vs sw.

We've only measured an 8% difference between Xcode 5 and Xcode 7 on the A9 -- I'm not sure where you're seeing 70% variance?

Also, the Xcode 7 was a one-off build to see how much Xcode has improved since we released 3.0.0. Geekbench 3 in the iTunes App Store will continue to be built with Xcode 5.0.0. Our policy is that we only upgrade compilers between major updates (e.g., 3.0.0, 4.0.0, etc). All minor updates (e.g., 3.0.1) use the same compiler.

 

[jfpoole]

FWIW, here's a comparison between the mobile and desktop workload sizes on the A9:

http://browser.primatelabs.com/geekbench3/compare/3594692?baseline=3594646

Mobile is on the left, desktop is on the right.

 

[Space69]

Thank you for participating.

We've only measured an 8% difference between Xcode 5 and Xcode 7 on the A9 -- I'm not sure where you're seeing 70% variance?

According to your own test - Sharpen filter etc.

Also, the Xcode 7 was a one-off build to see how much Xcode has improved since we released 3.0.0. Geekbench 3 in the iTunes App Store will continue to be built with Xcode 5.0.0. Our policy is that we only upgrade compilers between major updates (e.g., 3.0.0, 4.0.0, etc). All minor updates (e.g., 3.0.1) use the same compiler.

Will GB4 include any SIMD processing?
Since GB4 will use Apple Xcode 7 for IOS, will desktop include the latest Intel Compiler?

 

[Space69]

FWIW, here's a comparison between the mobile and desktop workload sizes on the A9:

http://browser.primatelabs.com/geekbench3/compare/3594692?baseline=3594646

Mobile is on the left, desktop is on the right.

Is the weighing between the 30 tests different when you compare mobile and desktop workloads?

 

[Eug]

Another useless browser JavaScript benchmark:

Jetstream 1.1:

152.48: Core i7 870 iMac 2.93 GHz running Safari 8.0.8 and Mac OS X 10.10.5
146.85: Core i7 870 iMac 2.93 GHz running Firefox 41 and Mac OS X 10.10.5
144.82: Core i7 870 iMac 2.93 GHz running Chrome 45 and Mac OS X 10.10.5

84.211: Athlon II X3 435 2.9 GHz running Firefox 41 and Windows 7 SP1

Error: Core 2 Duo T8300 MacBook 2.4 GHz running Safari 6.1.6 and Mac OS X 10.7.5
83.148: Core 2 Duo T8300 MacBook 2.4 GHz running Firefox 41 and Mac OS X 10.7.5
84.655: Core 2 Duo T8300 MacBook 2.4 GHz running Chrome 45 and Mac OS X 10.7.5

75.968: Core 2 Duo P8400 MacBook Pro 2.26 GHz running Safari 8.0.8 and Mac OS X 10.10.5
81.175: Core 2 Duo P8400 MacBook Pro 2.26 GHz running Firefox 41 and Mac OS X 10.10.5
81.403: Core 2 Duo P8400 MacBook Pro 2.26 GHz running Chrome 45 and Mac OS X 10.10.5

50.080: Core Duo T2500 iMac 2.0 GHz running Chrome 34 and Mac OS X 10.6.8
Script timeout: Core Duo T2500 iMac 2.0 GHz running Firefox 41 and Mac OS X 10.6.8

119.47: iPhone 6s with iOS 9.1 Public Beta 2

84.256: iPad Air 2 with iOS 9.1 Public Beta 2

57.389: iPhone 5s with iOS 9.1 Public Beta 2

21.140: iPhone 5 with iOS 9.1 Public Beta 2

Crash: iPad 2 with iOS 9.1 Public Beta 2

For the iPhone 6s, it probably should be a little higher than 120 because I didn't turn off screen auto-lock, so I had to keep tapping the screen to keep it awake.

tl;dr:

Out of all the bazillion computers and iDevices I have in my house, the iPhone 6s handily beats everything except my Core i7 870 in (the largely single-threaded?) JetStream 1.1.

 

[catinthefurnace]

:thumbsup:

Do you yet have anything to add to your earlier post about how the fact that Geekbench's memory score scales linearly with memory performance proves that Geekbench is a flawed benchmark that favors Apple SoCs, or is it just thumbs ups from here on out?

 

[catinthefurnace]

FWIW, here's a comparison between the mobile and desktop workload sizes on the A9:

http://browser.primatelabs.com/geekbench3/compare/3594692?baseline=3594646

Mobile is on the left, desktop is on the right.

Ok so if anything the A9 performs better using the desktop size workloads. What's it going to take for these conspiracy theorists to capitulate?

Is there anything? Or, are we just spinning our wheels as usual? Has everyone in this world lost the ability to change their mind in the face of evidence?

 

[Thanatosis]

Do you yet have anything to add to your earlier post about how the fact that Geekbench's memory score scales linearly with memory performance proves that Geekbench is a flawed benchmark that favors Apple SoCs, or is it just thumbs ups from here on out?

Why would a resident troll have anything to actually contribute besides misinformation and emoticons? That guy posts nothing but intel marketing slides and straight up lies, I don't know why he's tolerated. I guess because he tows the party line it's all good.


On topic, I really would like to see a few more people run geekbench on their A9 devices because I have seen two different sets of scores: 2300/4000 and 2550/4450. I think one of them is TSMC 16 FF and the other is SS 14nm FF (not sure which is which).

 

[catinthefurnace]

Why would a resident troll have anything to actually contribute besides misinformation and emoticons? That guy posts nothing but intel marketing slides and straight up lies, I don't know why he's tolerated. I guess because he tows the party line it's all good.

Sorry Thanatosis, I'm new to the forums here, and wasn't aware of his/her history.

 

[DrMrLordX]

That's a property of every code. So nothing to see here.

Unless you use something like Java; however, that shifts the performance burden onto the VM. There's a big difference in Oracle's JRE8 and Dalvik. Plus anything in Java is going to be a no-go on Apple's mobile hardware.