Apple A8 and A8X

Hasn't Apple always been optimistic about scaling? In that case that'd mean 3 cores @ 1.3GHz: 2x1.4Ghzx1.4= 3.92 = 3x1.3Ghz... That'd be very disappointing.

 

Too optimistic increase claims are common place for CPUs in general and not just for Apple. Other than that I don't see why they'd clock lower as in the iPad Air.

 

Why would they need 3 cores?

Could the software be reading the M8 co-processor as an extra core or?

 

Would seem very strange of Apple not to mention adding another CPU core. Given that they didn't have a lot of "reveals" for the ipad Air2 launch, the first apple soc with 3 CPUs would have been a relatively big deal for the launch.

 

"Big deal" is also highly relative to the fact that they also didn't mention how many clusters the GPU this time has.

 

I'm not sure so many people are tuned into GPU clusters. Enough for them to say x2.5.

However more people are used to talking about CPU cores. Now in a world of quad-core and octa-core, perhaps Apple doesn't want to be hit by "what your new soc only has 3 CPU cores", so maybe this is the reason.

But given Apple's pretty damn good CPU implementations, it won't be hard for it to be spun as apple being able to outperform octa-core devices, using "just" 3 cores.

Still seems a bit of a marketing miss, if true.

 

We'll have to wait for confirmation, but while triple core is nice, If that claimed 40% improvement was per core it would have been far more impressive.

Exactly this. They weren't afraid to trumpet the clock speed of the original iPad when it hit 1ghz but now that they've been left behind on that front they don't mention it. Same with the dual core A5. It had a core count at least equal to that of its contemporaries. Triple core is not a big deal these days in world of quad, hexa and octa cores.

 

A 3rd core would be a weird choice. Given that the iPad Air 2 has a smaller battery than the model it replaces, yet claims a similar battery life, you'd expect perf/w to have gone up, alongside improved display power consumption.

If Apple added a 3rd Cyclone core, which are wide and able to burn quite a few watts if unrestrained, then Apple would have to run the cores at a lower freq than in the iPhone 6, in situations where more than 1 core is active, potentially leading to times where the iPad is slower than the iPhone.

Is iOS 8 now a much better OS in terms of multi-tasking, that a 3rd core is a significantly better choice for the majority of iOS workloads than running two Cyclone cores at a much high freq than in the iPhone 6?, either option is going to affect power consumption.

 

Pretty sure it was a device improvement.

 

I doubt that the third core is significantly weaker than the other two because if it were then the bar on the PassMark page probably wouldn't be as long as it is.

I wonder if the supposed increases in core count and RAM (Hans thinks 4 GB, PassMark's Memory Mark chart says ~2 GB) are designed partly with the rumored iOS split-screen multitasking in mind.

Phil Schiller during the Apple presentation did give a 2x value that "some apps can achieve." He didn't mention any number greater than +25% during the A8's introduction last month, so the 2x number may not be for apps that happen to show large improvements on the A8 over the A7, but seems to lend weight to a third core or at least further per-core improvements in the A8X.

 

The PassMark's I/O benchmark is showing a 300+% increase for that iPad 5,4 device, compared to the iPhone 6 Plus, which is some jump.

http://www.iphonebenchmark.net/diskmark_chart.html

 

Yet many initial A8 rumors pointed at "quad core GPU" (irellevant if I dismissed them as nonsense).

3 is still better than 2 LOL

Consumption increases only if you constantly run all cores at full tilt; we both know that cases where hypothetically all 3 cores (if they are 3 after all) would run at their peak frequency should be extremely rare.

Is there really anything "unrestrained" these days in the ULP mobile space? It's all a matter of thread scheduling and DVFS. If things would be different then consumption for 4*A15+4*A7 big.LITTLE configs at around 2.0GHz frequencies for Android mainstream devices would shoot out of the roof. NV's 4+1 config isn't my favourite efficiency example, but while the minimum frequency of a whatever config containing A7 cores could start at 500MHz in NV's 4+1 the companion core goes as low as 50+MHz as its starting point.

I'd be seriously worried if you'd tell me that the cores don't run in an aSMP fashion.

 

Which was the previous X SOC, was it the A5X?

It was a costly piece of silicon wasn't it because of the size?

Maybe they decided to put all their baskets on the iPad Air 2 this time around -- as well as probably prioritizing iPhone 6 production.

Because obviously the iPad mini 3 got short shrift.

 

A5X was 165mm2 and A6X after that 123mm2; only A7 was a Vidal Sasoon/2in1 SoC for both iPhones and iPads.

If an A8X alike or even Tegra K1 SoC would be acceptable material for a form factor like smartphones we'd have some already. Above a certain threshold it's too much power consumption to tolerate for a smartphone.

As I said they could have skipped that one entirely.

On another note A8X being already at 3b transistors doesn't sound like there's THAT much mileage for SoCs to go in the foreseable future; eventually process problems are going to grind their teeth there also sooner than later.

 

Then again, from a purely architectural standpoint, being able to drop legacy 32-bit ISA support in their core designs from 10nm on or so would be really nice. Efficiency gains all around.

 

Using the 64-bit instruction set isn't always a win as far as performance goes due to increased branch prediction pressure (due to removal of conditional code execution) and data cache thrashing (due to 64-bit pointers). The latter can be somewhat mitigated by using ILP32, but I don't think Apple supports that.

It remains to be seen if the area gained by not supporting 32-bit code could compensate that loss...

 

Of course. Nevertheless I find the 64-bit ISA neater, and if you are not going to run any legacy code, being able to focus on AArch64 alone should simplify the design/debug process of new cores as well. No insult to the old ISA necessarily implied.

 

I definitely agree, I like the 64-bit ISA a lot more. I would even go as far as saying I'd like to see the 32-bit ISA die as soon as possible to concentrate on AArch64

 

Impressive ipad air2 geekbench results :

http://browser.primatelabs.com/geekbench3/1061742