Apple A6X

[runawayprisoner]

Surprisingly underscored in light of the new iPad Mini and what-nots.

So I'm wondering... do we even care that this SoC may well have the fastest ARM cores coupled with the fastest mobile graphics processing solution on the market?

2x graphics performance compared to A5X, which has yet to be beaten, is quite a jump in my opinions. It seems like we are approaching last-generation console quality...

But now, here's what I'm curious about: what sort of dope did Apple feed the GPU in order to make it 2x faster than a PowerVR SGX543MP4!?

 

[s44]

Exynos 5250 (new Chromebook + Manta) has the fastest ARM cores -- true A15.

Apple probably just doubled the GPU units.

 

[bearxor]

Surprisingly underscored in light of the new iPad Mini and what-nots.

So I'm wondering... do we even care that this SoC may well have the fastest ARM cores coupled with the fastest mobile graphics processing solution on the market?

2x graphics performance compared to A5X, which has yet to be beaten, is quite a jump in my opinions. It seems like we are approaching last-generation console quality...

But now, here's what I'm curious about: what sort of dope did Apple feed the GPU in order to make it 2x faster than a PowerVR SGX543MP4!?

The only thing I could think of, that I posted in the other thread, was a SGX543MP6 with a clock speed bump. Basically same thing they did with the iPhone 5. Add 50% more GPU, give it a clock speed bump to make up the rest. But wouldn't that be just a crazy amount of space?

Also, why? I mean, the 543MP4 was driving the 2048x1536 screen just fine. Did we really need double the GPU power for the new iPad? Couldn't we have just run into next year when we'll likely have a 600 series GPU paired with the A6?

I don't know of anyone that was going, "Man, I'd love to have an iPad 2 or 3, but the GPU performance is just HORRIBLE!"

 

[cmdrdredd]

How will games run on the huge resolution? That's what I'm wondering.

 

[Suspicious-Teach8788]

How will games run on the huge resolution? That's what I'm wondering.

If the A5X did fine, don't you think the A6X will?

 

[runawayprisoner]

Exynos 5250 (new Chromebook + Manta) has the fastest ARM cores -- true A15.

Apple probably just doubled the GPU units.

We'll see. The iPhone 5 was double the CPU performance of the 4S, but it wasn't exactly double the iPad 3. So there is still a chance Apple clocked the CPU higher in the A6X. Higher clocks would likely offset the performance benefits the A15 cores have over Apple's custom cores... (Anand calls the custom core Swift)

The only thing I could think of, that I posted in the other thread, was a SGX543MP6 with a clock speed bump. Basically same thing they did with the iPhone 5. Add 50% more GPU, give it a clock speed bump to make up the rest. But wouldn't that be just a crazy amount of space?

Also, why? I mean, the 543MP4 was driving the 2048x1536 screen just fine. Did we really need double the GPU power for the new iPad? Couldn't we have just run into next year when we'll likely have a 600 series GPU paired with the A6?

I don't know of anyone that was going, "Man, I'd love to have an iPad 2 or 3, but the GPU performance is just HORRIBLE!"

Yeah, I thought about a 543MP6, but that's a hell of a lot of space for one chip. I don't think that's likely at all.

But then... what else can they really do to make it 2x faster than the A5X? I don't know... maybe increase clock speeds even more?

It's evident from the iPhone 5 that Apple's A6 still has room to spare in terms of clock speed. Also considering the A6 didn't really beat A5X CPU performance by 2x, I'd think A6X will also feature a slight clock speed bump at the least.

So maybe it's the same 543MP4 but clocked even higher than what Apple put for the 543MP3 in the A6. That should give them the 2x figure.

How will games run on the huge resolution? That's what I'm wondering.

The iPad 3 was no slouch. It was mostly limited by RAM. 1GB (shared between CPU and GPU) isn't enough to push 2048 x 1536. If the A6X had 2GB RAM, then that'd be the bigger update. But damn... if it has 2x the GPU performance, we are seriously staring a beast in the face. It probably would chew up for snacks.

 

[A5]

If the A5X did fine, don't you think the A6X will?

I think most games were doing upscaled 1024x768, but yeah. The A6X will be fine.

 

[runawayprisoner]

I think most games were doing upscaled 1024x768, but yeah. The A6X will be fine.

No. They were upscaling from 1600 and up resolutions. Some at 1920 x 1200.

The limiting factor for the A5X was RAM. Apple had 4x more pixels but only 2x the amount of RAM.

 

[dagamer34]

Apple's website says the iPad 4 has "quad core graphics", so the 2x performance likely comes from increased clock speed of the PowerVR SGXMP4. Also, the memory subsystem of the CPUs have been improved, so it doesn't need to be a straight doubling of GPU clocks to hit a 2x performance increase.

 

[runawayprisoner]

Apple's website says the iPad 4 has "quad core graphics", so the 2x performance likely comes from increased clock speed of the PowerVR SGXMP4. Also, the memory subsystem of the CPUs have been improved, so it doesn't need to be a straight doubling of GPU clocks to hit a 2x performance increase.

It might actually take 2x GPU clocks. Or even more.

The A6 has SGX543MP3 running at 333MHz to match SGX543MP4 at 200MHz in the A5X.

That's just to "match". To really double the performance, it'd seriously take more than that even if you are taking into account the improved memory subsystem.

 

[Ravynmagi]

The limiting factor for the A5X was RAM. Apple had 4x more pixels but only 2x the amount of RAM.

Is the memory in the iPad 4 still 1GB?

 

[jpeyton]

We'll see. The iPhone 5 was double the CPU performance of the 4S, but it wasn't exactly double the iPad 3. So there is still a chance Apple clocked the CPU higher in the A6X. Higher clocks would likely offset the performance benefits the A15 cores have over Apple's custom cores... (Anand calls the custom core Swift)

Exynos 5250 is putting out sunspider scores around 600ms; A6 was around 900ms. A6X would need a good sized clock-speed boost to match.

 

[dagamer34]

It might actually take 2x GPU clocks. Or even more.

The A6 has SGX543MP3 running at 333MHz to match SGX543MP4 at 200MHz in the A5X.

That's just to "match". To really double the performance, it'd seriously take more than that even if you are taking into account the improved memory subsystem.

You'd only the need the SGX543MP4 clocked at 400Mhz to be 2x performance. We'd probably see PowerVR 6 series GPU in the A7.

 

[runawayprisoner]

Is the memory in the iPad 4 still 1GB?

That's what I'm wondering about.

Exynos 5250 is putting out sunspider scores around 600ms; A6 was around 900ms. A6X would need a good sized clock-speed boost to match.

Exynos 5250 runs at 1.7GHz while A6 is running at 1.3GHz. I think that's a good indication of clock speed advantage.

If Apple scales A6X enough, I think it'll match Exynos 5250 more or less... while offering much better battery life (Samsung Chromebook quotes 6.5 hours of run time).

You'd only the need the SGX543MP4 clocked at 400Mhz to be 2x performance. We'd probably see PowerVR 6 series GPU in the A7.

If performance still scales linearly at 2048 x 1536, then sure, that might be the case.

 

[ChronoReverse]

Exynos 5250 runs at 1.7GHz while A6 is running at 1.3GHz. I think that's a good indication of clock speed advantage.

Obviously way simplified but the difference between 1.3GHz and 1.7GHz is significantly smaller (in proportion) than the difference between 600ms and 900ms.

From this, it seems likely the A6 likely is closer to the S4 in terms of IPC than the A15. All of which stomp the A9 into the ground but that's beside the point.

 

[runawayprisoner]

Obviously way simplified but the difference between 1.3GHz and 1.7GHz is significantly smaller (in proportion) than the difference between 600ms and 900ms.

From this, it seems likely the A6 likely is closer to the S4 in terms of IPC than the A15. All of which stomp the A9 into the ground but that's beside the point.

A6 is not closer to S4. S4 in the HTC One X AT&T version shows a Sunspider score of roughly 1600ms.

So you're saying 1600ms to 900ms is not a significant difference but somehow 900ms to 600ms is?

That's to say nothing of A6 running at 1.3GHz compared to S4 running at 1.5GHz...

 

[ITHURTSWHENIP]

I dont put much stock into sunspider regardless of who is currently leading in it. CPU wise Cortex A15 is faster than Swift and Mali T-604 is rated at 72 gflops. Even if A6X wins its by a hair and definately not the same lead as last year

 

[ChronoReverse]

A6 is not closer to S4. S4 in the HTC One X AT&T version shows a Sunspider score of roughly 1600ms.

So you're saying 1600ms to 900ms is not a significant difference but somehow 900ms to 600ms is?

That's to say nothing of A6 running at 1.3GHz compared to S4 running at 1.5GHz...

I had forgotten about the S4's surprisingly low showings in Sunspider. They're particularly weird because the results in every other benchmark are much much better. In fact, the Snapdragon S4's Sunspider score is only marginally better than my Galaxy Note's Snapdragon S3 score.

With that said, Sunspider is a terrible benchmark since it varies tremendously depending on the OS and browser being used. For example, the same S4 chip as in the OneX scores 1300ms in the Motorola Atrix HD instead of 1600ms


Still, I based my comments on the fact the S4's single thread performance is usually more than 1.5 that of the A9 which puts it squarely equivalent the IPC increase when you compare the iPhone4S and the iPhone5. That is, when you equalize the clocks, the iPhone5's score is 1.5x the iPhone4.

Since the A15 is expected (and so far lives up to it) be faster than the S4 per clock, it seems that overall (outside of Sunspider), my original comment is still true.

 

[runawayprisoner]

Except that it's evident Exynos 5250 isn't going to win any power consumption award.

I'm doubtful we'll see A15 in any tablet or phone device for a while.

Meanwhile, A6 and A6X are already available in tablets.

Also rated GFLOPS is not the same as actual performance. On that same topic:
http://www.anandtech.com/show/6148/samsung-announces-a15malit604-based-exynos-5-dual

Anand had this to say about Mali-T604:

Depending on clock speeds I would expect peak performance north of the PowerVR SGX 543MP2, although I'm not sure if we'll see performance greater than the 543MP4

It's clear that 2x faster than Exynos 4 isn't going to make Mali-T604 competitive since A5X is already leading Exynos 4 by that much of a margin.

 

[cmdrdredd]

If the A5X did fine, don't you think the A6X will?

Fine by who's standards and what games are you talking about? Angry birds isn't enough...I need to see something more graphically intensive than what's out there now. Shadowgun on Tegra 3 uses physics and does not on the iPad. Whether this is due to using Nvidia proprietary physx or a limitation of the iPad I have no idea.

Anyway...the thing is, a lot of games on the PC run quite slow on a 2560x1440 resolution which is as close to 2048x1536 as I know of. Maybe we'll never see what I am hoping we start seeing on these tablets.

 

[ITHURTSWHENIP]

Considering ARM themselves claim Mali T-604 is 5x Mali 400 then im going out on a limb and say Anand is wrong. It wouldnt be the first time

We will find out monday regardless

 

[runawayprisoner]

I had forgotten about the S4's surprisingly low showings in Sunspider. They're particularly weird because the results in every other benchmark are much much better. In fact, the Snapdragon S4's Sunspider score is only marginally better than my Galaxy Note's Snapdragon S3 score.

With that said, Sunspider is a terrible benchmark since it varies tremendously depending on the OS and browser being used. For example, the same S4 chip as in the OneX scores 1300ms in the Motorola Atrix HD instead of 1600ms

Still, I based my comments on the fact the S4's single thread performance is usually more than 1.5 that of the A9 which puts it squarely equivalent the IPC increase when you compare the iPhone4S and the iPhone5. That is, when you equalize the clocks, the iPhone5's score is 1.5x the iPhone4.

Since the A15 is expected (and so far lives up to it) be faster than the S4 per clock, it seems that overall (outside of Sunspider), my original comment is still true.

I think it's not that A15 is expected to be faster than S4 that's the problem. It's obvious Apple's Swift core is also faster than S4 where it counts. Problem: S4 isn't the performance beast we all expected it to be.

If we take A9 as the baseline, then Swift is actually either as fast as A15 or faster than A15. Why? Because ARM could only claim a 40% performance improvement over A9 for A15.

http://www.itproportal.com/2011/03/14/exclusive-arm-cortex-a15-40-cent-faster-cortex-a9/

Apple's Swift showing 50% performance improvement over A9 clock-by-clock is actually quite awesome.

Considering ARM themselves claim Mali T-604 is 5x Mali 400 then im going out on a limb and say Anand is wrong. It wouldnt be the first time

We will find out monday regardless

ARM writes "up to", not "definitely". I think the performance advantage roughly depends on how many cores are configured with the SoC, as well as the clock speed.

Unless Samsung runs Mali T-604 at its absolute max performance configuration possible, I don't think you'll see actual performance reaching that point. In which case, Anand is not completely off base.

 

[ChronoReverse]

I think it's not that A15 is expected to be faster than S4 that's the problem. It's obvious Apple's Swift core is also faster than S4 where it counts. Problem: S4 isn't the performance beast we all expected it to be..

But it's not. There's only the Sunspider benchmark which I've already shown is a terrible benchmark.

In every other sort of test, INCLUDING user experience tests (as opposed to the extremely synthetic Sunspider), S4 has been stomping all over A9.

 

[ViRGE]

It might actually take 2x GPU clocks. Or even more.

The A6 has SGX543MP3 running at 333MHz to match SGX543MP4 at 200MHz in the A5X.

That's just to "match". To really double the performance, it'd seriously take more than that even if you are taking into account the improved memory subsystem.

I think you're forgetting just how mediocre A9's memory controller was. It would be very reasonable for a SGX543MP4 at 333MHz(ish) and coupled with Swift's memory controller (expanded to 128bits) to double A5X's graphics performance in certain situations. Primarily those raster/ROP bound as opposed to shader bound.

 

[ChronoReverse]

Just to hammer home how useless Sunspider is as a benchmark, the Microsoft Surface, running a 1.3GHz Tegra 3 (A9) gets 969 milliseconds in Sunspider.

An A9.