Apple A9X the new mobile SoC king

[coercitiv]

Pre orders will start in october, for some reason I thought they had allowed pre-orders at the same time as the 6s and they just weren't shipping until october, in which case I would have placed an order for one with the 6s.

Then what did you pick up for the wife?

 

[Thanatosis]

Then what did you pick up for the wife?

Same thing I got but a different color: Rose Gold 64GB 6s. I thought I had ordered the ipad pro with it (and would have, had it been available) but I guess I was recalling something else. I'll have to order the 128GB LTE iPad Pro in october.

 

[Phynaz]

We need a backpedal emoticon.

 

[Thanatosis]

We need a backpedal emoticon.

Why so IDC can use it every time he posts? :biggrin:

 

[stingerman]

I wonder just how much faster A10 will be over A9 (from a combination of IPC and clock speed).

With faster transistors, they can have a deeper and wider pipeline with additional multipliers, etc. Significantly more instructions in flight and higher instruction level parallelism (ILP). Along with a higher clock speed. And there is one more feature which would seem natural for such a processor to significantly increase ILP.

Simultaneous multithreading (SMT). Using the past processor wars as a lesson, it's better for Apple to implement SMT instead of competing for the highest core count. And, it's significantly more power efficient to add SMT. My conservative estimate is that we'll see 30% but I feel comfortable they'll get a 50% boost over the A9, especially if SMT is added.

 

[stingerman]

And don't forget the rumors include using TSMCs InFO-WLP which stacks components vertically and efficiently connects them together. So Apple could be condensing the size of their logic board significantly, like the S1 in the Apple Watch. This should also significantly reduce power requirements in addition to space. The iPhone 7 should be quite a beautiful phone.

 

[Arachnotronic]

And don't forget the rumors include using TSMCs InFO-WLP which stacks components vertically and efficiently connects them together. So Apple could be condensing the size of their logic board significantly, like the S1 in the Apple Watch. This should also significantly reduce power requirements in addition to space. The iPhone 7 should be quite a beautiful phone.

If true, that would be quite compelling.

 

[soccerballtux]

The problem is that you won't know if Apple doesn't do specific benchmark optimizations.

no matter what tire you install, if it's spinning at 100 RPM, you're still only going 70mph

 

[Arachnotronic]

With faster transistors, they can have a deeper and wider pipeline with additional multipliers, etc. Significantly more instructions in flight and higher instruction level parallelism (ILP). Along with a higher clock speed. And there is one more feature which would seem natural for such a processor to significantly increase ILP.

Simultaneous multithreading (SMT). Using the past processor wars as a lesson, it's better for Apple to implement SMT instead of competing for the highest core count. And, it's significantly more power efficient to add SMT. My conservative estimate is that we'll see 30% but I feel comfortable they'll get a 50% boost over the A9, especially if SMT is added.

At first this seems overly optimistic but it may not be.

I think they can get higher frequency with the A10 over the A9, esp. since it seems to be the case that the 16FF+ process is more efficient than 14nm LPP. A boost to ~2GHz seems reasonable (+10%) especially as 16FF+ will have had time to mature.

From a ST perf/clock perspective, I think another 10% seems reasonable. If SMT is added, this could lead to 20-30% improved performance in MT loads.

So A9 * 1.10 (frequency boost) * 1.10 (perf/clock boost) ~= 21% ST performance improvement.

Layering SMT on top of that could give a best case of 30% uplift (assuming negligible perf/thread degradation from addition of SMT which I would say is optimistic), leading to 57% more performance in a best-case, blue-sky scenario.

 

[zentan]

It seems Idontcare have been the most consistent, with what he has said earlier.
Just a humble request to Idontcare,don't stop posting because of some wayward comments.Your insights are quite good to read.

 

[EightySix Four]

A user poll seems to show the TSMC phones getting about 2 hours longer battery life.

http://www.engadget.com/2015/10/06/does-your-iphone-have-a-good-or-bad-a9-cpu/

 

[erunion]

A user poll seems to show the TSMC phones getting about 2 hours longer battery life.

http://www.engadget.com/2015/10/06/does-your-iphone-have-a-good-or-bad-a9-cpu/

Interesting. I've been interested in how Apple was going to dual source SOC.
They could have the foundries fab SOCs for different products(eg 6s vs 6s+) but then they wouldn't have flexibility to change the product mix.

Or they could put different chips into the same product, which is what they did.

Its also interesting that redditers assumed the Samsung chip would be superior based solely on die size. The fact that Apple chose TSMC for the A10 should have suggested otherwise.

 

[EightySix Four]

Interesting. I've been interested in how Apple was going to dual source SOC.
They could have the foundries fab SOCs for different products(eg 6s vs 6s+) but then they wouldn't have flexibility to change the product mix.

Or they could put different chips into the same product, which is what they did.

Its also interesting that redditers assumed the Samsung chip would be superior based solely on die size. The fact that Apple chose TSMC for the A10 should have suggested otherwise.

I assumed it would as well, not just based on die size, but the fact that they won Nvidia's business as well. Others have reported that running the two side-by-side, Samsung has a slight but consistent advantage in benchmarking applications. Maybe a bit faster (which Nvidia cares about) at the cost of power. It will be interesting to see if these results could be replicated in a controlled environment - if so I imagine a lot of the Samsung supplied phone users will be pissed.

 

[DrMrLordX]

if so I imagine a lot of the Samsung supplied phone users will be pissed.

Hmm . . . one wonders if owners of iPhones with the Samsung-sourced A9s will request replacements with TSMC-sourced chips? What kind of battery life does Apple currently (or at any point up until now) advertise for the device?

 

[Thanatosis]

Hmm . . . one wonders if owners of iPhones with the Samsung-sourced A9s will request replacements with TSMC-sourced chips? What kind of battery life does Apple currently (or at any point up until now) advertise for the device?

I have a samsung iPhone 6s, I won't be returning it for another because I have had no problems with it in normal use. It appears to have better battery life than either my ipad air 2 or my old iphone 5s.


Maybe people with TSMC phones have 14+ hour wifi browsing but the ~11hrs I'm getting is pretty consistent with what Apple advertised, so meh.


I think it's sort of foolish to freak out about it, it would be like getting a 4770k and returning it because it only overclocked to 4.1Ghz. Some people would return their CPU for that, though, I know.... so it's up to the user I guess.

 

[Arachnotronic]

Hmm . . . one wonders if owners of iPhones with the Samsung-sourced A9s will request replacements with TSMC-sourced chips? What kind of battery life does Apple currently (or at any point up until now) advertise for the device?

I have a Samsung-built A9...not going to return it because frankly it's too much of a bother.

Looking forward to getting a good TSMC-built chip with the iPhone 7 next year though :thumbsup:

 

[lopri]

Does anyone know what Geekbench battery test measures and how to interpret the results? My Xperia Z Ultra's runtime is 4 hours 45 minutes and Galaxy S6's runtime is 7 hours. Both give me about ~5 hours screen-on-time in day to day use.

 

[lothar]

I have a samsung iPhone 6s, I won't be returning it for another because I have had no problems with it in normal use. It appears to have better battery life than either my ipad air 2 or my old iphone 5s.


Maybe people with TSMC phones have 14+ hour wifi browsing but the ~11hrs I'm getting is pretty consistent with what Apple advertised, so meh.


I think it's sort of foolish to freak out about it, it would be like getting a 4770k and returning it because it only overclocked to 4.1Ghz. Some people would return their CPU for that, though, I know.... so it's up to the user I guess.

But many people here do that all the time.
See D0 stepping of i7 920 and Pentium D for reference.

If one is okay with returning an i7 920 that they fired up in their new rig for 2 minutes only to realize that they didn't get the D0 stepping that they wished for, why wouldn't they do the same for an iPhone?

 

[stingerman]

I think they can get higher frequency with the A10 over the A9, esp. since it seems to be the case that the 16FF+ process is more efficient than 14nm LPP. A boost to ~2GHz seems reasonable (+10%) especially as 16FF+ will have had time to mature.

The A9X should be interesting with what it reveals, especially if there is a significant Mhz bump.

 

[Eug]

Apple says real world differences in battery life are 2-3%.

http://techcrunch.com/2015/10/08/ap...rmance-of-new-iphones-a9-chips-vary-only-2-3/

 

[jhu]

Apple says real world differences in battery life are 2-3%.

http://techcrunch.com/2015/10/08/ap...rmance-of-new-iphones-a9-chips-vary-only-2-3/

Seems like people want to make mountains out of molehills. If you're making a compute cluster using iPhone 6, then you'll see this issue. Also you might want to rethink using phones for a compute cluster.

 

[erunion]

Seems like people want to make mountains out of molehills. If you're making a compute cluster using iPhone 6, then you'll see this issue. Also you might want to rethink using phones for a compute cluster.

I disagree. Apple admitted that the results were genuine. They then went on to say the test isn't representative of real usage.

But the difference might be significant for someone who isn't the 'typical' user.

 

[cytg111]

404

 

[tempestglen]

I disagree. Apple admitted that the results were genuine. They then went on to say the test isn't representative of real usage.

But the difference might be significant for someone who isn't the 'typical' user.

Normal usage, 2-3% difference, but if you test geekbench continously, the difference is quite big.

 

[DrMrLordX]

Maybe people with TSMC phones have 14+ hour wifi browsing but the ~11hrs I'm getting is pretty consistent with what Apple advertised, so meh.

If the battery life is consistent with what Apple advertised for the product, then consumers will have no place to complain about some of the devices performing better-than-advertised while others perform as expected.