Cortex A-15 v Bulldozer/Piledriver

[DavidC]

DAVID C; Thanks for the breakdown, so from your numbers if i have read it correct, BD is 20% faster than Athlon xp??
If so that does sound reasonable and pretty much confirms what i thought looking at all those benchies...

No, Athlon X2 is a dual core Athlon 64, which is 20-30% faster than Athlon XP.

 

[french toast]

No, Athlon X2 is a dual core Athlon 64, which is 20-30% faster than Athlon XP.

Well, in that case i revise my assumptions then...Athlon 64 is within 10% of Bulldozer...with A15 being 30% slower than BD and 20% off Athlon 64..

I site these benchies as suporting evidence;
http://www.anandtech.com/bench/Product/403?vs=90

As already stated, Husky is deemed to be around 5-10% faster than BD..and this test clearly shows the older Athlon keep up with and beat Husky in some cases..when the Husky cores clearly are favoured as they are running on a newer platform.

How do you breakdown those tests?

 

[DavidC]

Well, in that case i revise my assumptions then...Athlon 64 is within 10% of Bulldozer...with A15 being 30% slower than BD and 20% off Athlon 64..

French Toast, the Athlon X2 7850 is based on K10/Phenom core(notice the L3 cache). The fastest chip based on the true X2(which are dual core versions of Athlon 64) is the Athlon X2 6400+. Sysmark benefits from quad cores, so its hard to compare that to Bulldozer which has 8. and X2 is a dual core.

Here, comparison of A6-3650 with similarly clocked Athlon X2: http://www.anandtech.com/bench/Product/403?vs=32
You can see that the X2 7850 is noticeably faster than the X2 5200+ at the same clocks: http://www.anandtech.com/bench/Product/78?vs=32

 

[french toast]

French Toast, the Athlon X2 7850 is based on K10/Phenom core(notice the L3 cache). The fastest chip based on the true X2(which are dual core versions of Athlon 64) is the Athlon X2 6400+. Sysmark benefits from quad cores, so its hard to compare that to Bulldozer which has 8. and X2 is a dual core.

Here, comparison of A6-3650 with similarly clocked Athlon X2: http://www.anandtech.com/bench/Product/403?vs=32
You can see that the X2 7850 is noticeably faster than the X2 5200+ at the same clocks: http://www.anandtech.com/bench/Product/78?vs=32

Ok ok, fair enough, you convinced me with that last link that they are different Micro architectures,...Athlon 64 is noticebly slower than K10..

The other comparison (first) is actually a very good one, that might well be the best representation we have of A15 v PD (in x86 world)..

Heres my thinking; Piledriver will increase IPC to around Husky Levels...the A6-3650 is clocked, cached, threaded, and likely ddr 3 make it the 'Piledriver' side of the equation..agree??

The X2 5300+ is right around the ball park where A15 is going to be..give or take.
If you take a look at that comparison...and level the playing field some more..ie...up the core count to 4...quadruple the L2.....slap on some ddr3.... then the gap would close significantly to A6-3650.. as like you say SYSmark is multithreaded/ likely sensitive to cache also.

So if we take consideraion into the above and narrow the scores somewhat..we might be looking at a decent/approx/vague representation of Cortex A15 v Piledriver IPC comparison.
http://www.anandtech.com/bench/Product/403?vs=32
You agree?

 

[DavidC]

I think Piledriver will get to Deneb levels, not quite Husky. Ok, maybe in some scenarios that take advantage of the L3 cache that Husky doesn't have. I can't really say about the others.

 

[french toast]

Well ive just noticed something that will make all the difference...Assuming we go off Bulldozer and not Piledriver. (as PD is going to be another 30% faster) then what may level the playing field is whether you were to benchmark both setups off mains power OR battery...if you go with the latter that is going to definately favour ARM...

How much? just check out these graphs measuring a Intel i5-2537m...the battery scores point to a massive drop in performance.. Sunspider gets 2240..thats actually beaten by Cortex A9's currently? Krait@ 1.5ghz pulls in 1600 odd?...im not suggesting for minute that a cortex A9 is anywhere near a Sandy bridge part..however @17W? food for thought...
http://techreport.com/articles.x/21551/6

 

[swaaye]

The numbers there from Sunspider look strange in general. Maybe it's a different version of the bench? I see Firefox 3.6.9 was used, which could be the issue because Firefox used to be very slow with JS.

Here's the Tech Report bench run of Sunspider from another review. The test methodology page says they used Chromium v9 (Chrome for Linux).
http://techreport.com/articles.x/21813/11

Android devices have stock browsers based on Webkit, and Chromium/Chrome is based on Webkit too.

 

[french toast]

Yea the software is old...but that linux version you ran doesn't show the 17w version running off battery...i think im going to dig around for some other results

 

[swaaye]

It's just important to note that javascript performance is a moving target on both x86 and Android because the web browsers are continually improving.

 

[DavidC]

Here it shows that battery life difference between High Performance and Battery mode is only 10%, yet the High Performance mode is 3.3x faster in Javascript. In Balanced mode it would perform within 5% of the High Performance and close all the battery life gap from Battery.

http://techreport.com/articles.x/21902/8

The battery mode probably runs it at LFM, which is 800MHz, without any Turbo.

 

[french toast]

Yea the only thing that can be compared is the processors running the exact same software, in this case on that list, maybe the actual numbers them selfs are not imporyant, more the difference between battery/no battery per architecture....at least in this 17w i5m case not good:smile:

Found another example;

http://techreport.com/articles.x/21902/6

 

[DavidC]

No one will run that on that mode though, look at page 8.

 

[swaaye]

I ran Sunspider on some old hardware today.

Mobile Athlon 64 3000+ (1.8 GHz / 1024K L2), 2GB PC2700 single channel, XP SP3 [eMachines M6805]
490.2ms (Firefox 11)
766.8ms (Chrome 18)

Celeron M 900 MHz (512K L2), 2GB PC2-3200 single channel, XP SP3 [EeePC 900]
986.9ms (Firefox 11)
1340.0ms (Chrome 18)


It seems impossible to use these review charts to compare CPUs because if you don't run the same browser the comparison is meaningless.

Also, I am surprised how much better Firefox scores than Chrome!

 

[DavidC]

Also, I am surprised how much better Firefox scores than Chrome!

That's strange, I get almost the same results between the two:

Sunspider Javascript 0.9.1

Firefox 11: 188.4ms
Chrome 18: 186.5ms

Fastest is IE9

153.3ms

 

[swaaye]

Maybe it's XP related? I even turned off CnQ on the A64.

Both machines had fresh XP SP3 installs. The A64 has a discrete Radeon 9600 while the EeePC uses GMA 900.

 

[Blazkowicz]

waiting for such SunSpider benchmarks where ARM and PC CPUs are compared, running the same version of the browser.

you only have to run a full version of chromium or firefox on an ARM device, use debian or ubuntu if needed but Android might cut it if it runs a genuine full version.

btw anyone installed Android on an x86 PC?, that would be a fun experiment.

 

[swaaye]

I've run the various versions of Android x86 on my EeePC 900. ASUS is part of the project and there are specific EeePC ISO images available. It runs very smoothly but navigating Android and its apps by touchpad doesn't work as well as by touch. Also, it's not possible to hide the soft button bar on Android 3/4 and get true full-screen in most cases.

Unfortunately in order to run it I have to swap out my wifi board because I replaced the original Atheros board with an Intel 5100 card and Android x86 doesn't have drivers for it. Not worth the effort to replace just for some benching because I have to fully disassemble the EeePC.

 

[Tahir2]

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone/


ARM A15 (Snapdragon S4) vs X86 Medfield

Dual vs single core with HT

1.5GHz vs 1.6GHz

Adreno 225 vs PowerVR SGX 540

The Snapdragon holds it own in most benchmarks - some of the GPU benchmarks are not apples for apples as there is a resolution difference between the two phones.

Android on X86 is running via binary translation, some apps that are native coded rather than through Dalvik won't run at all (Adobe Flash 11 and Netflix mentioned in the review.)

IMHO - ARM has some serious competition but right now still is the better solution.

How is this related to this topic? If ARM A15 is being pegged at times by the Intel Medfield SoC already - then expect Bulldozer/Piledriver/Bobcat/Jaguar to be faster still. Just need those processors in the same TDP range now!

 

[ninelven]

Just for clarity S4 != A15.

 

[Tahir2]

Just for clarity S4 != A15.

Sorry, are you asking or informing?

AFAIK the Qualcomm Snapdragon S4 is based on the A15 architecture. Qualcomm design their own cores under license from ARM.