Apple A5X SoC

[dagamer34]

Well, we are going to get 3DMark for Windows 8, which will level performance comparisons for non-Apple chips, since I assume that most of the SoCs will eventually be on Windows 8 tablets. But then we run into more potential bias minefields from people who think 3DMark is a waste of time and we should only bother testing games (of which, none bother having an FPS counter on by default).

So while benchmark applications are not ideal, sites like AnandTech have to work with what they can, short of writing a custom benchmark for themselves.

 

[runawayprisoner]

I think the problem here is that people just can't accept that Tegra 3 is not the best SoC on the market... Bias or not. Even nVidia said that they would test Apple's claim personally last week, and we haven't heard anything back from them since then. I'm quite certain that there is a chance nVidia actually conducted a test of their own and found that Apple's claim was quite accurate.

Sincerely, if Tegra 3 is that powerful, I'd think Sony would have used it for the Vita instead, since Tegra 3 is likely more power efficient than the power-gobbling combo that Sony put in the Vita.

On paper, even PowerVR SGX543MP2 in A5 has more raw performance number, so it shouldn't be a surprise that A5X would be faster now, should it?

As for the bias of the benchmark, sincerely, I don't think there will ever be a benchmark that isn't considered biased if you think GLBenchmark is biased because there is only PowerVR on iOS' side, and iOS in its entirety is made to run on PowerVR. In which case, the iOS version of the benchmark is always made to run on a PowerVR chip, regardless of which benchmark it is.

 

[Steelbom]

1024x768 on a 10" screen already looked pretty good. Combine that with the easy scaling, and I really doubt any developers will use the full resolution for high-end games.

Why's that? Real Racing 2, Galaxy On Fire 2 HD are both running at the higher resolution, and Sky Gamblers: Air Supremacy, Modern Combat 3, and such, all look to be running at the higher resolution too.

 

[BenSkywalker]

I keep reading GLBenchmark favoring PowerVR architecture, yet I'm not sure I "know" about something like that. Mind enlightening me?

The GLBench results everyone likes to use are the onest that render to an off screen buffer. You know what PowerVR chips draw that is off screen? *Nothing*. That is a central element to their design. The scene data would get processed, nothing gets drawn.

I'm quite certain that there is a chance nVidia actually conducted a test of their own and found that Apple's claim was quite accurate.

That's funny, Mac fan sites found them not even close to true, actually the A5X lost most of the benches straight up. Guess they must have simply been building up a reputation as Mac sites for years just so they could come out as biased at this exact moment when compared against a SoC, brilliant viral marketing by nV I'm sure

BTW- PowerVR chips have been used on Android phones too, I pointed out the same problems with GLBench back then. Difference is, the Android guys tend to use things like logic and simply look at how shockingly bad the bench is on its' own merit.

 

[MrX8503]

Does it really matter?

The MP4 is clearly faster than the Tegra 3, not 4x, but still a lot faster. Also it looks like games perform perfectly fine at native res...so that prediction about games looking horrible turned out false. Why are we still complaining again?

 

[cheezy321]

Does it really matter?

The MP4 is clearly faster than the Tegra 3, not 4x, but still a lot faster. Also it looks like games perform perfectly fine at native res...so that prediction about games looking horrible turned out false. Why are we still complaining again?

Because you are an apple fan and you don't use logic!

 

[runawayprisoner]

The GLBench results everyone likes to use are the onest that render to an off screen buffer. You know what PowerVR chips draw that is off screen? *Nothing*. That is a central element to their design. The scene data would get processed, nothing gets drawn.

Any article detailing this behavior? And shouldn't other GPUs do the same?

That's funny, Mac fan sites found them not even close to true, actually the A5X lost most of the benches straight up. Guess they must have simply been building up a reputation as Mac sites for years just so they could come out as biased at this exact moment when compared against a SoC, brilliant viral marketing by nV I'm sure

BTW- PowerVR chips have been used on Android phones too, I pointed out the same problems with GLBench back then. Difference is, the Android guys tend to use things like logic and simply look at how shockingly bad the bench is on its' own merit.

I haven't seen you "point out the same problem". And as far as benches go, Tegra 3 mostly won in CPU benchmarks, so why is that relevant to the graphics performance in question again?

 

[Steelbom]

The GLBench results everyone likes to use are the onest that render to an off screen buffer. You know what PowerVR chips draw that is off screen? *Nothing*. That is a central element to their design. The scene data would get processed, nothing gets drawn.

And the developers of GLBenchmark couldn't force it to draw off screen?

Does it really matter?
The MP4 is clearly faster than the Tegra 3, not 4x, but still a lot faster. Also it looks like games perform perfectly fine at native res...so that prediction about games looking horrible turned out false. Why are we still complaining again?

It is actually 4x faster. The MP2 in the iPad 2 scores 1.5x higher than the Tegra 3's GPU in GLBenchmark 2.1 Egypt 720p, and 2x higher in GLBenchmark 2.1 Pro 720p. The latter is more texture bound, and it's where the PowerVR SGX543MP2 does very well, and the iPad 3 has twice as many, an SGX543MP4, so performance should be doubled. The scores at 720p for the iPad 3 are probably bottlenecked in these tests though.

 

[runawayprisoner]

Well, Anand's results are in. Quad-channel memory! Who woulda thunk?

And performance at 2048 x 1536 isn't so bad, either. Looks like this'll be the benchmark for mobile GPU performance for a while.

 

[Mopetar]

Well, Anand's results are in. Quad-channel memory! Who woulda thunk?

And performance at 2048 x 1536 isn't so bad, either. Looks like this'll be the benchmark for mobile GPU performance for a while.

Probably until next year's iPad chip. I really don't see any other manufacturer being interested in making such a large chip or one that generally has a lot of needless graphics power.

 

[BenSkywalker]

It is actually 4x faster.

The results are in, overall the A5x gets stomped pretty badly by the Tegra3 in CPU tasks, narrowing it down to the GPU the 4x results don't seem to show up for any of the benches. In one of the GLBench results they show the A5x to be ~3x faster, but some of the primitive tests in GLBench show Tegra3 to be 5x-7x faster then the A5x.

Also, the GPU doesn't have enough power to run games at the native resolution based on what Anand stated. I'm sure something like AngryBirds would run just fine, but more demanding games seem like they are getting scaled.

Side note- nice that Apple managed to get close to the level of contrast offered by the hyper expensive Kindle Fire with their new wunder display.....

 

[MrX8503]

I read all the way through Anands review and it was a good read. The CPU performance remains the same because it's the same CPU but the GPU performance is impressive. It looks like Apple achieved what they were set out to do which is maintain performance at a 1536p res. Performance disparity will likely increase with the Asus infinity with its increased res at 1080p while still using a tegra 3.

Anand seemed to be impressed that Apple was able to fit the iPad with such a quality display at a $499 price point. I agree with his sentiments that this may pave the way for better displays for laptops and tablets in the future. Hopefully for laptops as laptop displays are awful.

We may see retina MacBooks this year, if not, it'll happen eventually as there already HiDPI settings in Lion. Definitely exciting for SOCs and the displays it powers.

 

[Puddle Jumper]

Well, Anand's results are in. Quad-channel memory! Who woulda thunk?

And performance at 2048 x 1536 isn't so bad, either. Looks like this'll be the benchmark for mobile GPU performance for a while.

Not necessarily, Windows 8 should lead to Ivy Bridge and Trinity powered tablets which would easily dethrone A5X.

 

[MrX8503]

Ivy bridge tablets? I think that still constitutes as "a while".

 

[Puddle Jumper]

Ivy bridge tablets? I think that still constitutes as "a while".

To be honest it wouldn't even take Ivy Bridge. Even an AMD Brazos APU should be able to beat A5X on the GPU front.

 

[badb0y]

The results are in, overall the A5x gets stomped pretty badly by the Tegra3 in CPU tasks, narrowing it down to the GPU the 4x results don't seem to show up for any of the benches. In one of the GLBench results they show the A5x to be ~3x faster, but some of the primitive tests in GLBench show Tegra3 to be 5x-7x faster then the A5x.

Also, the GPU doesn't have enough power to run games at the native resolution based on what Anand stated. I'm sure something like AngryBirds would run just fine, but more demanding games seem like they are getting scaled.

Side note- nice that Apple managed to get close to the level of contrast offered by the hyper expensive Kindle Fire with their new wunder display.....

 

[theeedude]

Chipworks posted an iPad3 wallpaper sized pic:

It's a huge die, and they are wasting some area between partitions.

 

[smartpatrol]

Not necessarily, Windows 8 should lead to Ivy Bridge and Trinity powered tablets which would easily dethrone A5X.

LOL you'll be waiting a while for that. The first Windows 8 x86 tablets will be based on Intel Clover Trail -- their dual-core Atom-based SOC. Oh, and hope you're willing to pay over $600 for them:
http://www.engadget.com/2012/01/16/intel-windows-8-tablet-pricing/

 

[bearxor]

Probably until next year's iPad chip.

I don't think so. For Apple to go any farther, they would need to move to the Series 6 GPU, and I don't think those will be ready for integration until Q2 '13 at the earliest. I don't think moving to a 544 or 554 makes any sense. I suppose they COULD add more cores, but I don't know if it's necessary.

I think the A6 chip shows up in the iPhone (2012) with a DC A15 @ 800mhz or so and the same 543MP2. The A6X shows up in iPad (2013) with DC A15 @ 1ghz and the 543MP4. Then, MAYBE the iPhone (2013) gets a Series-6 GPU. Maybe.

 

[Puddle Jumper]

LOL you'll be waiting a while for that. The first Windows 8 x86 tablets will be based on Intel Clover Trail -- their dual-core Atom-based SOC. Oh, and hope you're willing to pay over $600 for them:
http://www.engadget.com/2012/01/16/intel-windows-8-tablet-pricing/

You're forgetting about AMD's Brazos platform which is their competitor for Atom but with substantially more powerful AT HD 6000 series gpus.

Price isn't that big of a concern for me provided the hardware justifies it.

 

[smartpatrol]

You're forgetting about AMD's Brazos platform which is their competitor for Atom but with substantially more powerful AT HD 6000 series gpus.

Price isn't that big of a concern for me provided the hardware justifies it.

Brazos looks promising. I'm excited to see how it stacks up against the ARM competition.

I'm not holding my breath though. Yeah, x86 and Radeon are far more powerful than the mobile SOCs, but it remains to be seen how well they can perform when scaled down to the ~2w range. If there's one thing I've learned from following the CPU/GPU industries, it's that nothing is a sure bet until it's released. If you are holding out hope for a white knight, you may very well be disappointed (and I'd hate to say it, but this ESPECIALLY applies to AMD as of late).

 

[MrX8503]

To be honest it wouldn't even take Ivy Bridge. Even an AMD Brazos APU should be able to beat A5X on the GPU front.

I think you're ignoring their power draw. Ivy and Brazos aren't going to be making it into tablets.

 

[Puddle Jumper]

I think you're ignoring their power draw. Ivy and Brazos aren't going to be making it into tablets.

The iPad 3 proved that tablets could support higher power consumption than we previously thought. If they used an iPad 3 sized battery with a more conventional screen the lower power AMD APUs should have no problem delivering acceptable battery life in a tablet.

 

[alent1234]

The iPad 3 proved that tablets could support higher power consumption than we previously thought. If they used an iPad 3 sized battery with a more conventional screen the lower power AMD APUs should have no problem delivering acceptable battery life in a tablet.

apple took the hit one year because they can't buy 32nm and better CPU's in the quantities they are asking for. next year expect apple's CPU's to be A LOT more power efficient

 

[MrX8503]

The iPad 3 proved that tablets could support higher power consumption than we previously thought. If they used an iPad 3 sized battery with a more conventional screen the lower power AMD APUs should have no problem delivering acceptable battery life in a tablet.

I don't think its going to be that simple. How are they going to cool it? How are they going to support both a power hungry CPU and a high resolution display? I don't know the TDP of the A5X, but I don't think its 7W like the display is. Even if they were able to get this to work, A5X is here and now, an AMD APU alternative would be anywhere from months to a year away. Knowing AMD, probably never.

Seems like its a lot more work than its worth. It would be easier to go with ARM or use Intel Medfield.