Apple A6X

[runawayprisoner]

I already said in my first post that its entirely possible that A6X is faster (GPU wise) but not by the same lead A5X has enjoyed over its competitors. You are the one who has taken an absolute stance, not me.

As for Cortex A15 vs Swift, the 40% increase ARM claimed is for DMIPS alone. Samsungs paper says 1,7 GHz Cortex A15 is 2X faster than a 1,4 GHz Cortex A9

You were saying that it's only "by a hair".

I agree that it may not be the same lead, but at the current rate, it's likely 3/4 the performance of A6X, and that's still quite a lot when you start graphing things. Not to mention Mali T-604 won't be available on any phone or tablet this Christmas season.

It's also to note that Apple improved a number of things moving from A5 to A6, so I don't think it'll simply be a double frequency A5X GPU. In that case, the stakes are much higher.

But all of that aside, A6X will at least reign as the supreme SoC for this Christmas season. And I think that's what counts... rather than having to wait all the way until Spring or Summer.

I'll hold off on A15 vs Swift until we get more concrete benchmark scores for A15. It's really hard to tell now.

 

[s44]

Not to mention Mali T-604 won't be available on any phone or tablet this Christmas season.

Phone yes. Tablet -- el Goog isn't announcing a device on Monday so they can tease it for three months, you know.

 

[runawayprisoner]

Phone yes. Tablet -- el Goog isn't announcing a device on Monday so they can tease it for three months, you know.

You mean the Nexus 10? Something tells me El Goog may use Snapdragon S4 Pro instead of Exynos 5250 for that device.

The telling indication being that Exynos 5250 only scores 6.5 hours of usage on a bigger Chromebook. Unless El Goog wants to release a tablet that lasts 40% less than an iPad, I think they want a more power efficient SoC.

If Google ends up using S4 Pro, it's not because Exynos 5250 isn't a good performer, but that it's terrible for devices needing high battery life.

 

[Skurge]

You mean the Nexus 10? Something tells me El Goog may use Snapdragon S4 Pro instead of Exynos 5250 for that device.

The telling indication being that Exynos 5250 only scores 6.5 hours of usage on a bigger Chromebook. Unless El Goog wants to release a tablet that lasts 40% less than an iPad, I think they want a more power efficient SoC.

If Google ends up using S4 Pro, it's not because Exynos 5250 isn't a good performer, but that it's terrible for devices needing high battery life.

That chromebook's battery isn't that much bigger than an ipad's is it?

 

[Mopetar]

Some time in 2013. We'll probably see the first announcements at CES, though it's anyone's guess how long after that it will be until those products actually ship.

Probably as good of a guess as any. Another question is who's going to be using it other than Apple and Intel? nVidia obviously uses their own, Qualcomm uses their own (Adreno), Samsung uses the ARM-made Mali GPUs, and TI has said that they're selling off their SoC business so they probably won't be making anything with it.

 

[runawayprisoner]

That chromebook's battery isn't that much bigger than an ipad's is it?

Which iPad? The 3rd gen iPad needs a bigger battery because the Retina Display needs more backlight in order to push the same brightness to your eyes.

But it still lasts 10 hours.

 

[s44]

You mean the Nexus 10? Something tells me El Goog may use Snapdragon S4 Pro instead of Exynos 5250 for that device.

Not possible. Adreno 320 only does 20*12, not 25*16. Only the new Mali does 25*16.

Not sure what the deal is with battery. Clocks, keyboard, OS? We'll see soon enough.

 

[lothar]

Probably as good of a guess as any. Another question is who's going to be using it other than Apple and Intel? nVidia obviously uses their own, Qualcomm uses their own (Adreno), Samsung uses the ARM-made Mali GPUs, and TI has said that they're selling off their SoC business so they probably won't be making anything with it.

Who are they selling it to?

 

[Mopetar]

Who are they selling it to?

I don't know if anyone has been officially announced as the buyer, but I recall reading an article stating that Amazon had expressed some interested in order to have their own chips for their Kindle products. If no one wants to buy them out, I'd assume that they just lay off the team and possibly close office space. I honestly haven't read much more into the situation beyond what was posted on these forums within the last few weeks, so I couldn't say if there are any actual interested buyers beyond Amazon or even how interested Amazon is in such a proposition.

 

[ITHURTSWHENIP]

You were saying that it's only "by a hair".

I agree that it may not be the same lead, but at the current rate, it's likely 3/4 the performance of A6X, and that's still quite a lot when you start graphing things. Not to mention Mali T-604 won't be available on any phone or tablet this Christmas season.

It's also to note that Apple improved a number of things moving from A5 to A6, so I don't think it'll simply be a double frequency A5X GPU. In that case, the stakes are much higher.

But all of that aside, A6X will at least reign as the supreme SoC for this Christmas season. And I think that's what counts... rather than having to wait all the way until Spring or Summer.

I'll hold off on A15 vs Swift until we get more concrete benchmark scores for A15. It's really hard to tell now.

If a 1,5 GHz T-604 is 3X Mali 400 and SGX543MP*500 MHz is 4x Mali-400 then it stands to reason that even if there is zero driver improvements then A6X will be less than 0.9x more powerful than a 1,7 GHz Mali-T604. Compare that to A5X wich had a 2-3x lead over everyone else.

Its not like any of this matters. Even if Exynos 5 will be in the Nexus 10 on monday and obliterates iPad in benchmarks, Apple will sell 10 iPad 4s for every Nexus 10 at that price.

 

[runawayprisoner]

If a 1,5 GHz T-604 is 3X Mali 400 and SGX543MP*500 MHz is 4x Mali-400 then it stands to reason that even if there is zero driver improvements then A6X will be less than 0.9x more powerful than a 1,7 GHz Mali-T604. Compare that to A5X wich had a 2-3x lead over everyone else.

Its not like any of this matters. Even if Exynos 5 will be in the Nexus 10 on monday and obliterates iPad in benchmarks, Apple will sell 10 iPad 4s for every Nexus 10 at that price.

Nope. 1.7GHz Mali T-604, even assuming linear scaling, is still only 13% faster than 1.5GHz. Considering 4x is 33% faster than 3x, the difference between A6X and 1.7GHz Mali T-604 is still roughly 20%, not less than 10%.

A5X only led Mali 400 by 100%, not 200% or 300%... unless you are looking at triangle fill and texture fill benchmarks, but then that means the gap between A6X and Mali T-604 is even larger in those benchmarks.

We'll see, but I don't think the difference is as small as you'd believe.

And yeah, it doesn't matter. But then the discussion is mostly to see what we can expect from A6X, and not about how many iPads Apple can sell.

Last year, Apple more or less matched expectations for A5X, but I was off by a huge margin regarding the CPU (no improvement over A5).

 

[Skurge]

Nexus 10 is coming out with Exynos 5250 so it should be available around the same time the ipad 4 will. So they can be compared.

 

[runawayprisoner]

Yeah. Let's see how Nexus 10 does.

iPad 4:

Looks like a minor 100MHz clock bump as opposed to a full bump to 1.6GHz. Either that or Geekbench is again doing that "sorry, we read clock speeds wrong" stunt again.

Also 1GB RAM. Looks like it'll run into the same memory problems the iPad 3 did... again. Not sure why Apple didn't bump memory up to 2GB.

 

[ChronoReverse]

Looks like the S4's poor Sunspider performance in Android is likely a software issue (not as well optimized).

Anand's test of the HTC 8X got 914ms . Does terrible in Octane and Browsermark so lol software optimizations.

They really need to hack Linux onto all the devices and run a native bench.

 

[Zink]

I think we should just throw out Sunspider now. It is only useful for comparing devices running the same OS and even then OEMs do their own optimizations and software versions are different so you never know.

I was kind of surprised that the reported speed for A6X is only 1.4 GHz but that sore is only 7% higher than the iPhone 5 so it seems right. Apple claims A6 is up to 2x faster than A5 and also claim A6X is up to 2x A5X. I was expecting 1.6 GHz or more given 800 MHz to 1300 Mhz on the phone. There's a big battery in there and 28 nm saves idle power so I was expecting a bit more juice.

 

[dagamer34]

I think we should just throw out Sunspider now. It is only useful for comparing devices running the same OS and even then OEMs do their own optimizations and software versions are different so you never know.

I was kind of supervised that the reported speed for A6X is only 1.4 GHz. Apple claims A6 is up to 2x faster than A5 and also claim A6X is up to 2x A5X. I was expecting 1.6 GHz or more given 800 MHz to 1300 Mhz on the phone, they have the battery capacity to do it I would have though?

No, it shows that most of the gains in the A6 and A6X are largely from improvements made to areas other than raw CPU architecture, like the memory subsystem which is said not to be that great in the Cortex-A9.

 

[bearxor]

I think we should just throw out Sunspider now. It is only useful for comparing devices running the same OS and even then OEMs do their own optimizations and software versions are different so you never know.

I was kind of surprised that the reported speed for A6X is only 1.4 GHz but that sore is only 7% higher than the iPhone 5 so it seems right. Apple claims A6 is up to 2x faster than A5 and also claim A6X is up to 2x A5X. I was expecting 1.6 GHz or more given 800 MHz to 1300 Mhz on the phone. There's a big battery in there and 28 nm saves idle power so I was expecting a bit more juice.

The A5 an the A5X are the same, processor wise. The A6 and A6X are (presumably) the same, processor-wise.

 

[Zink]

The A5 an the A5X are the same, processor wise. The A6 and A6X are (presumably) the same, processor-wise.

Same CPU but the iPad has always been clocked at 1 GHz instead of around 800 MHz on the phone. A6 and A6X score within 7% of each other in Geekbench so much closer than before but they still claim 2x for both, maybe the iPhone is a bit over and the iPad hits it right on in whatever metric they are using.

 

[Zink]

Much faster than before. If this isn't even a new architecture, I wonder what will happen next generation with completely a new GPU architecture in there.
http://www.glbenchmark.com/compare....=Apple iPad 4&D2=Apple iPad 3&D3=Apple iPad 2

 

[gus6464]

Looks like the S4's poor Sunspider performance in Android is likely a software issue (not as well optimized).

Anand's test of the HTC 8X got 914ms . Does terrible in Octane and Browsermark so lol software optimizations.

They really need to hack Linux onto all the devices and run a native bench.

Also since the Surface hits 900's on Sunspider with a pretty weak CPU compared to what's out there by Samsung and Qualcomm.

 

[Mopetar]

Also 1GB RAM. Looks like it'll run into the same memory problems the iPad 3 did... again. Not sure why Apple didn't bump memory up to 2GB.

What memory problems?

I've got 6 browser tabs open, have been downloading and listening to several podcasts, have used a few different apps that download or stream video, and GTA III is still loaded in memory and immediately ready to go from where I closed out of it even though I haven't played it five days.

Compared to the first generation iPad where having more than two browser tabs open would probably cause the others to start reloading, the iPad 3 doesn't have that problem for me.

More memory would obviously be better, but I really can't see the need for it. Maybe your workload is different than mine though.

 

[Aikouka]

Anand's A6X GPU Preview: http://www.anandtech.com/show/6426/ipad-4-gpu-performance-analyzed-powervr-sgx-554mp4-under-the-hood

 

[runawayprisoner]

If a 1,5 GHz T-604 is 3X Mali 400 and SGX543MP*500 MHz is 4x Mali-400 then it stands to reason that even if there is zero driver improvements then A6X will be less than 0.9x more powerful than a 1,7 GHz Mali-T604. Compare that to A5X wich had a 2-3x lead over everyone else.

Its not like any of this matters. Even if Exynos 5 will be in the Nexus 10 on monday and obliterates iPad in benchmarks, Apple will sell 10 iPad 4s for every Nexus 10 at that price.

And the jury is out...

Like I "thought", Mali T-604 is still no match at all. In fact, at native resolution (2560 x 1600), it's not even a match for PowerVR SGX543MP4 in some cases.

What memory problems?

I've got 6 browser tabs open, have been downloading and listening to several podcasts, have used a few different apps that download or stream video, and GTA III is still loaded in memory and immediately ready to go from where I closed out of it even though I haven't played it five days.

Compared to the first generation iPad where having more than two browser tabs open would probably cause the others to start reloading, the iPad 3 doesn't have that problem for me.

More memory would obviously be better, but I really can't see the need for it. Maybe your workload is different than mine though.

It's a problem when you start using more memory intensive apps... like image editors, or painting/drawing apps. Some games like Infinity Blade II also suck memory like crazy. GTA III is actually on the light side due to lower resolution textures.

 

[MrX8503]

And the jury is out...

Like I "thought", Mali T-604 is still no match at all. In fact, at native resolution (2560 x 1600), it's not even a match for PowerVR SGX543MP4 in some cases.




It's a problem when you start using more memory intensive apps... like image editors, or painting/drawing apps. Some games like Infinity Blade II also suck memory like crazy. GTA III is actually on the light side due to lower resolution textures.

Heh, I wonder where those people are claiming the Mali would wipe the floor of the 4th gen iPad.

There you have it, the iPad is a beast.

 

[runawayprisoner]

Heh, I wonder where those people are claiming the Mali would wipe the floor of the 4th gen iPad.

Well, from careful observations of released data, I already deduced that T-604 would be about 3/4 that of A6X, which is just about right. (it's 50% slower in general)

It's surprising to see it fall behind by almost 100% or more in some benchmarks, though.

Also surprising is how Exynos 5250 in Nexus 10 performs almost 100% worse than in the Chromebook. Something tells me it's either thermal throttling (which would explain T-604 results as something similar happened to the Nexus 4 in those benchmarks compared to the Optimus G), or drivers optimization issues. I'm banking on the first option seeing as Cortex A15 does have higher power consumption (necessitating the BIG.little design) and that some benchmarks do have the Exynos 5250 pulling ahead (suggesting it's not really an optimization issue).

It would also explain why Apple decided to move to a custom core this generation. They knew Cortex A15 wouldn't be a viable solution for their thinner/lighter philosophy, and so they had to create an alternative that would not be thermal throttled that easily.