Apple A8 and A8X

3C@1.5GHz sounds about right for that 40% claim.

 

More than anything those memory scores are outstanding. Seems the CPU has full access to the 128bit interface.

Edit: Also notice the L2 cache is being reported as 2MB instead of 1MB on the A8 http://browser.primatelabs.com/geekbench3/compare/1061936?baseline=1061742

 

Compared to Denver K1: http://browser.primatelabs.com/geekbench3/compare/1037443?baseline=1061742

 

Denver is pretty good performer if you take in account transistor budget. Competition Is a major force drive innovation.

 

Any single process uses more memory in LP64 mode, but if all 3rd party processes move to LP64, overall system memory usage should be substantially reduced.

Today, if you have two active processes using different architectures, you keep the long tail of linked libraries in memory for both architectures. Once the last 32b process goes away, there is a significant step in reducing memory pressure.

 

But we don't know the transistor budget.

 

Well, that Denver SoC is also clocked 1Ghz faster.

Look at the floating point performance of the iPad5,4.

 

No L3 detected, but then again it wasn't for A7 either, so if any changes apart from what was seen with the A8 has been made we can't see it here. IPC is very impressive, and the low clock implies that clock rates could be fairly sustainable, judging by the performance of the similar iPhone6. That remains to be seen though,

 

It looks like the Geekbench 3 single-core CPU performance of A8X in iPad Air 2 is ~ 12% higher overall than than that of A8 in iPhone 6, with most of the difference due to a higher CPU clock operating frequency in iPad Air 2.

With Cyclone and enhanced Cyclone CPU, the Geekbench 3 dual-core score appears to be ~ 1.777x higher than the single-core score. In other words, for every doubling of cores, the Geekbench 3 multi-core score should increase by ~ 1.777x. Since there are three cores in iPad Air 2, the multi-core score is ~ 2.47x higher than the single-core score ( 2.47 = 1.777*(1+1.777)*0.5 ).

Hard to compare accurately with Denver because we don't have AArch64 results yet for that CPU core.

 

Overall integer performance (5015 v 1801) and overall float performance (4829 v 1698), are showing x2.78 and x2.84 improvement respectively from single core to tri-core.

the mem tests show a x1.3 improvement, which is reducing the overall difference between single and tri-core scores.

 

Engadget's performance numbers:

 

There's still quite a bit that can be fine tuned in their drivers it seems.

 

Is it the CPU Physics score that is still holding back A8X in 3dmark Ice Storm Unlimited overall score? The third CPU core should have significantly improved the Physics score. The Graphics score should be close to mobile Kepler, but based on a combined score of 21,659, it appears that the Graphics score is less than 30,000...?

 

Graphics: 31500
Physics: 10400

 

So graphics score is clearly quite good, but the Physics score is only slightly higher than A8. I was under the impression that 3dmark Unlimited Ice Storm Physics test was able to take good advantage of extra CPU cores (up to 4 cores)???

 

Compared to the iPhone 6/Plus's scores, the iPad Air 2 does 20-25% better in 3DMark IS Unlimited. That scaling is much worse than the other benchmarks and is bizarrely anemic for a chip that physically has 50% more of everything along with slightly higher clocks.

I wish that Futuremark would make post somewhere stating "this is what we do, how we do it, and we are 100% right". Right now, though, I'm quite willing to assume that 3DMark's code has a bottleneck that doesn't apply to any other game/benchmark.

 

No it doesn't have "everything" by 50% scaled, otherwise they'd still scale in cores and not clusters. What we are sure about for now is that a 6650 has 50% more ALUs & TMUs than a 6450 but there's quite a bit more a GPU consists of than just the former.

Other than that until I see the fillrate behaviour of the 6650 I can't estimate if the frequency is higher than on the 6450 in the 6 Plus.

What I can estimate right now from the very limited data provided is that the onscreen Manhattan score needs work; we had seen some weird early behaviour on GK20A in that regard also but it was quickly rectified. With the bandwidth the A8X has, the onscreen score sounds weird.

Honestly I myself mostly look at Gfxbench results but even there for more than one test I'm not completely sure what exactly the application is doing.

I had the impression that 3dmark's recent mobile graphics benchmarks are heavily bandwidth limited; either it's not true or there's still work to do (see above).

 

What numbers were you looking for instead? Comparing to the Shield Tablet and taking its resolution (1920x1200) into account, the Air 2's onscreen results look normal to slightly above expectations.

 

The first result is offscreen, the second onscreen (2048*1536)

iPad Air 2: 32.4/24.6fps
Shield tablet: 31.0/29.7fps

First K1 onscreen results in Manhattan were below 20 fps.

 

Nvidia's mid-2014 Tegra K1 Shield Tablet has a 1920x1200 screen.

Edit: I'll link to GFXBench's Shield Tablet page too.