Apple's A10X rumored to be a hex-core CPU

[Mondozei]

If we assume the rumor is correct, do you still think Apple isn't serious about moving beyond phones and tablets with their CPUs?

 

[Arachnotronic]

If we assume the rumor is correct, do you still think Apple isn't serious about moving beyond phones and tablets with their CPUs?

Utterly stupid rumor. The A9 is a dual core design. Why on Earth would Apple move from a sensible, high performance dual core design to a hex core design?

 

[Mondozei]

Utterly stupid rumor. The A9 is a dual core design. Why on Earth would Apple move from a sensible, high performance dual core design to a hex core design?

It could be A10X. Which would make sense if Apple continues to push into higher and higher performance markets, such as laptops. We also don't know how many cores A9X has right now, but if it's a quad core, going to hex isn't a big jump and makes sense if we're going to see even larger performance jumps.

2016 could well be the year when Apple goes all in on its own SoCs powering everything that they have.

 

[Arachnotronic]

It could be A10X. Which would make sense if Apple continues to push into higher and higher performance markets, such as laptops. We also don't know how many cores A9X has right now, but if it's a quad core, going to hex isn't a big jump and makes sense if we're going to see even larger performance jumps.

2016 could well be the year when Apple goes all in on its own SoCs powering everything that they have.

A9X is a triple core design. Apple has shown with its designs that it understands that "moar cores" in thermally constrained power envelopes really doesn't make a whole lot of sense.

 

[Mondozei]

A9X is a triple core design.

Source?

Apple has shown with its designs that it understands that "moar cores" in thermally constrained power envelopes really doesn't make a whole lot of sense.

The key assumption.

 

[NTMBK]

Perhaps it's tri-core with SMT? Sounds more like Apple's style.

 

[VirtualLarry]

Perhaps it's tri-core with SMT? Sounds more like Apple's style.

That I could believe. Someone else commented that Apple's A9 (A9X?) core is wide enough, that it's ripe for SMT to be added.

 

[raghu78]

That I could believe. Someone else commented that Apple's A9 (A9X?) core is wide enough, that it's ripe for SMT to be added.

yeah that was me. I believe Apple's custom ARMv8 core is definitely right up there with Intel's big core in terms of IPC and execution resources. Apple needs to get SMT implemented as its a massively wide core with lots of execution resources. 15-20% improvement in multi thread performance is pretty much guaranteed. Add to it that Apple keeps pushing IPC up by 5-10% every year and it makes all the sense to fully utilize such a massive core. The A10 could be 2C/4T and A10X could be 3C/6T. Such a design would be brilliant in both single and multithread performance.

 

[Eug]

If we assume the rumor is correct

Probably the smarter thing to do is to assume the rumour is incorrect.

 

[Mondozei]

yeah that was me. I believe Apple's custom ARMv8 core is definitely right up there with Intel's big core in terms of IPC and execution resources. Apple needs to get SMT implemented as its a massively wide core with lots of execution resources. 15-20% improvement in multi thread performance is pretty much guaranteed. Add to it that Apple keeps pushing IPC up by 5-10% every year and it makes all the sense to fully utilize such a massive core. The A10 could be 2C/4T and A10X could be 3C/6T. Such a design would be brilliant in both single and multithread performance.

Yeah makes sense. But what if they introduce a new A10 SoC. A10 for phones, A10X for tablets and A10Y for laptops or even desktop PCs and up.

They are already doing two derivatives. If they were serious about the laptop space, why not add one more derivative, in which a hexcore, or at least a quad core, would make sense.


P.S. I'm still waiting for someone to provide source for A9X being tri-core.

 

[Arachnotronic]

P.S. I'm still waiting for someone to provide source for A9X being tri-core.

Apple claimed +70% CPU performance in going from A8 to A9. We know this was at iso core count.

Apple claimed +80% CPU performance in going from A8X to A9X. Do you really think they added a fourth core when we know this gain can be achieved with a combination of frequency and perf/clock improvements?

 

[sm625]

Triple core with SMT I could believe. But 6 cores? No way.

 

[davygee]

Yeah makes sense. But what if they introduce a new A10 SoC. A10 for phones, A10X for tablets and A10Y for laptops or even desktop PCs and up.

They are already doing two derivatives. If they were serious about the laptop space, why not add one more derivative, in which a hexcore, or at least a quad core, would make sense.

P.S. I'm still waiting for someone to provide source for A9X being tri-core.

I see the sense in that. So Apple could potentially stick with 2 cores (iPhone), 3 cores (iPad) & 4 cores (MacBook).

If Apple manage to develop their chips on the same scale as previous years, could be seeing the A10 pushing 6-7k, A10X pushing 9-10k and an A10Y (4 core) easily upwards of 10k and possibly significantly higher if they can up the clockspeed.

We should know more once we see benchmarks of the A9X which should be in the 6000-8000 range.

 

[lopri]

Why would Apple go for SMT.. SMT makes sense when there is lots of inefficiency around execution.

 

[nAo^2]

If they want to replace x86 cores in their laptops they could use 1-2 extra cores to help with dynamic translation of x86 code to ARM8 + hot spot optmizations.

 

[Thanatosis]

Why would they implement a hexcore for what will ostensibly be the processor for the next iPad, when their current tri-core A9X is already competing directly with intel's flagship Skylake Core M devices? It would be massive overkill to have a chip scoring 10k+ multicore in a tablet.


That said, the current A8X and A9 chips already appear to be overkill for the current ipad software. Maybe they are really targeting full notebook, but wouldn't that require re-compiling the OSX kernel?

 

[Eug]

Why would they implement a hexcore for what will ostensibly be the processor for the next iPad, when their current tri-core A9X is already competing directly with intel's flagship Skylake Core M devices? It would be massive overkill to have a chip scoring 10k+ multicore in a tablet.

That said, the current A8X and A9 chips already appear to be overkill for the current ipad software. Maybe they are really targeting full notebook, but wouldn't that require re-compiling the OSX kernel?

A8X is definitely not overkill for all current iPad software. Even on an iPhone with A8, some software needs all the performance it can get. For example simple iMovie edits still take a while to render. This will only get worse with 4K video. Furthermore, numerous iPhone 6s reviewers state that the phone is noticeably faster than the iPhone 6 just for overall feel. This is even more true with the iPhone 6s Plus, where people complained about slowdowns even last year.

I'm personally hoping A10 for my desired iPhone 7 has a built in image processor that's fast enough to provide electronic image stabilization for 4K. This is currently not possible for the iPhone 6s. EIS is only supported up to 1080p. Sure, this is not traditional "CPU" performance but it just illustrates that these SoCs still need to evolve and improve.

 

[2is]

Well they now have the iPad Pro which would have the necessary thermal characteristics as well as the functionality to make use of the power. I don't think it's as far fetched as some of you seem to think it is. For the next iPhone yeah, not going to happen. For the next iPad Pro, it's certainly possible.

 

[Systems analyst]

Maybe Apple will implement 2-core SoCs on their phones, 3-core SoCs on the iPad, 4-core SoCs on the iPadPro, and 6-core SoCs for MacBooks. The possible reasons are as follows:
1. Apple seem to work very deliberately, and do not over-specify their devices; e.g. 3-core SoCs for iPad, not 4-core.
2. They need to differentiate their array of products, and the SoCs are one aspect of this.
3. There is thought to be a market for larger tablets, for genuine professional use, hence the work with IBM; this would need extra processing power, hence 4 cores.
4. ARM and partners estimate the market for AIOs and Laptops at 250 million/year in 2020; they are focussed on addressing this opportunity.
5. Apple are the most advanced SoC designer, and are certain, in my view, to want to address this market themselves, and not leave it to others.
6. Intel's 6th generation mobile CPUs, such as i7-6600U (4 thread, 15W) and i7-6700HQ (8 thread, 45W) are the standard established by Intel.
7. A 6-core Apple SoC would compete at the level of i7 mobile.
8. Multi-threading is probably ruled out by the need to be a power-miser, in the usual Apple tradition; ARM enlarged on this in a 2013 paper.
9. Costs would be considerably reduced, compared with Intel processors.
10. ARM are coming with A72 (Maia) powered devices on 14/16 nm in 2016.
11. ARM powered devices using ARTEMIS, the successor to A72, should arrive in 2017, and ARTEMIS should be targetted on 10nm.
12. The successor to ARTEMIS will arrive one year later in 2018, in devices.
13. Apple have to move more rapidly, to stay ahead of mass-market devices, using stock ARM IP.
14. TSMC 10nm may be in full production by 2017, maybe enabling a 6-core solution.

 

[Tuna-Fish]

Why would Apple go for SMT.. SMT makes sense when there is lots of inefficiency around execution.

When you have lots of execution resources, no matter how good your pipeline is you will have plenty of unused resources caused by dependencies between instructions. Introducing SMT allows you to capture (some of) these unused resources for relatively little cost.