Next-Gen iPhone & iPhone Nano Speculation

[ltcommander.data]

http://www.macrumors.com/2012/10/05/new-ipad-shows-up-in-app-analytics-running-a6-based-chip/

An A6-class 3rd gen iPad refresh may be in the works. A developer has started seeing an iPad3,5 show up using an ARMv7s based CPU. A true 4th gen iPad should be an iPad4,1 although you never know seeing the 2nd gen iPhone 3G was just the iPhone1,2. I guess the possibilities include this being an imminent iPad 3 refresh for this year, this being next year's $399 full size iPad to replace the iPad 2, this being next year's iPad 4 or this simply being a prototype that will never see the light of day. This isn't the iPad Mini since that seems to be the iPad2,5 and iPad2,6 that has previously shown up with A5 references.

If they do put out an iPad 3 refresh this year, I wonder if they will basically use the A6 as is or if it'll be a unique chip? With the A6 offering A5X equivalent GPU performance a direct substitution with maybe a minor clock speed bump would be sufficient for a refresh. They might miss the 128-bit memory controller at that iPad Retina resolution though. So perhaps an A6X with a return to the SGX543MP4 operating at A6 clock speeds (~325 MHz vs 250MHz) with a 128-bit bus and LPDDR2-1066? That would be good for a 30% GPU performance bump. Appropriate for fall 2012 iPad 3 refresh but a 2014 iPad 4 would need to be more aggressive. Apple churning out 3.5 SoC in a year (A5X, A6, A6X with the 32nm A5 being the half) would be quite a feat.

 

[wco81]

They claimed they couldn't meet demand of iPhone 5.

Last year, they said they could have sold more than the 37 million iPhone 4S that they reported for the holiday quarter.

I would think they're hoarding A6 and all the fab capacity to maximize iPhone 5 supply for the rest of the year.

 

[ltcommander.data]

They claimed they couldn't meet demand of iPhone 5.

Last year, they said they could have sold more than the 37 million iPhone 4S that they reported for the holiday quarter.

I would think they're hoarding A6 and all the fab capacity to maximize iPhone 5 supply for the rest of the year.

But is limited iPhone 5 supplies due to constrained A6 production/fab capacity? It seems the more commonly reported constraints are for in-cell touch screen production and QC problems with the housing. As such there may well be enough A6/32nm fab capacity for an iPad 3.5. The concern would be whether limited in-cell touch screen production means the iPad 3.5 would have to ship without one which would be disappointing for a flagship device considering even the new iPod Touch has one. However, it's interesting to note that Samsung doesn't seem to be a supplier for the touch screen in the iPhone 5 whereas they are the primary supplier of Retina displays for the iPad. It may well be that Apple is dedicating certain suppliers to each product line, which means that the iPad could get in-cell screens after-all since Samsung making them doesn't take away from existing iPhone 5 capacity.

An iPad 3 refresh now wouldn't make much sense if the iPad 4 is still going to be due in March 2013. However, an October 2012 iPad 3.5 might make sense if Apple wants the iPad 4 to have Rogue and OpenGL ES 3.0 to truly be a generational improvement. If Rogue isn't ready until closer to mid-2013, then the iPad 4 would have to be delayed from it's usual March release leaving the iPad 3 on sale another quarter. It wouldn't surprise me if Apple wanted to get rid of the iPad 3 as soon as possible since it's very much a compromise being thicker, heavier, and having less battery life than the iPad 2 in order to accommodate the Retina display and LTE and now lacks the Lightning connector. An October iPad 3.5 would then be well timed being 7 months after the iPad 3 and 8 months before a WWDC June release of the iPad 4. If the iPad 4 was coming in March 2013, a straight reuse of the A6 with little clock speed change for the iPad 3.5 would be most efficient. But if the iPad 4 isn't coming until June 2013, I think it'd make sense to do a dedicated A6X for the iPad 3.5, say 1.5GHz dual core ARMv7s with 375MHz SGX543MP4 and a 128-bit memory bus with LPDDR2-1066. That'd provide a 2x increase in CPU performance, a 50% improvement in GPU performance, and a 33% improvement in theoretical memory bandwidth over the A5X. And since only 1 GPU core is being added over the A6 the percentage die area increase will be smaller than the A5>A5X transition to help mitigate fab capacity concerns.

An October 2012 iPad 3.5 (with iPad Mini?) and June 2013 iPad 4 with the new iPhone and iPod Touch already announced will mean a long gap with no iOS-related announcements, but given the state of Maps they could probably focus on software and place an iOS 6.1 event during that time to highlight Maps, Siri, and iCloud improvements in order to maintain media attention on their ecosystem. Or maybe some long rumoured product could finally be completed and announced during that gap like an Apple television set or iTunes streaming service.

 

[wco81]

Just bought my iPad 3 in June or July.

Would be somewhat ticked if they came out with another iPad so soon, though unless there's a big jump in CPU/GPU, it won't necessarily obsolete my iPad so quickly, as far as apps. being updated to support higher performance profile (esp. if they put more RAM in the SOC).

Some speculation has been that Sharp IGZO displays would be the upgrade path to reduce power consumption and thereby reduce the thickness and weight of the form factor.

 

[dagamer]

Just bought my iPad 3 in June or July.

Would be somewhat ticked if they came out with another iPad so soon, though unless there's a big jump in CPU/GPU, it won't necessarily obsolete my iPad so quickly, as far as apps. being updated to support higher performance profile (esp. if they put more RAM in the SOC).

Some speculation has been that Sharp IGZO displays would be the upgrade path to reduce power consumption and thereby reduce the thickness and weight of the form factor.

We have to remember that the iPad 3 leaked several months in advance even though the only difference was a few millimeters increase in thickness. And leakers would certainly notice a new 9.7" iPad with the Lightning connector, it's unmistakeable.

Because of the smaller size of the Lightning connector, I don't think Apple can make a tiny revision to the device without being tempted to go all the way. Most of the refinements that an iPad 4 would have can't just be suddenly stuck into a hypothetical 3.5. Even when products have obvious shortcomings, Apple still waits for their yearly refresh in order to catch everything up, even delaying the product additional months compared to expectations in the case of the iPhone 4S so that everything launches on time.

What we can extrapolate however is that if/when the iPad Mini does get a retina screen, I don't ever see it using the 45nm A5X. That chip has way too many compromises. It's going to get a die shrink at some point so the 32nm version has identical performance as the 45nm version, and Apple doesn't have to worry about the loss of memory bandwidth performance that comes with using the A6.

 

[wco81]

Well the latest rumors are that Apple has ordered 10 million iPad Minis for the 4th quarter.

That's less than what the regular iPad would be expected to sell in the holiday quarter. Yet how much will iPad Mini cannibalize regular iPad sales?

Apple would ideally hope for none or minimal cannibalization. But you never know, iPad 3 will be one of the oldest tablets available this Holiday season, with Amazon undercutting them for the 10-inch config while there will be several W8 tablets that get a lot of attention as well.

 

[anexanhume]

Well the latest rumors are that Apple has ordered 10 million iPad Minis for the 4th quarter.

That's less than what the regular iPad would be expected to sell in the holiday quarter. Yet how much will iPad Mini cannibalize regular iPad sales?

Apple would ideally hope for none or minimal cannibalization. But you never know, iPad 3 will be one of the oldest tablets available this Holiday season, with Amazon undercutting them for the 10-inch config while there will be several W8 tablets that get a lot of attention as well.

I think this a product where you have little concern about cannibalization. It's about protecting the market. The only harm in cannibalization is likely going to be profit margins, but those are so robust elsewhere it shouldn't be a huge deal. The only question I have is how silly will the ipod touch prices look after the mini is priced?

 

[wco81]

I guess if the iPad Mini sells in the volumes they hope, while the iPod Touch 5G sales suffer, they can either bail on offering iPod Touch in the future or find a way to reduce costs on it to make it more competitive.

I think they'd be able to sell the 16 GB Touch for $249 in decent volumes, unless the market really is moving towards 7-inch tablets and "phablets" as alternatives to smart phones.

You can't pocket those as easily as the Touch but people are making do and making the tradeoff in size, weight and reduced battery life.

 

[ltcommander.data]

http://www.macotakara.jp/blog/ipod/entry-17953.html

Well benchmarks for the 5th gen iPod Touch are in. Looks like its A5 is identically clocked to the iPhone 4S (800MHz CPU/200MHz GPU). Teardowns probably won't take long to confirm the assumption it's using the 32nm version. I thought given the 18% increase in resolution there was a chance Apple would use a more iPad 2 like frequency (1 Ghz CPU/250MHz GPU) to better maintain the same GPU performance per pixel. I wonder if that means game developers will optimize separately for the 5th gen iPod Touch and iPhone 4S (not the just resolution difference but simplify effects for the iPod Touch) or just base the level of graphics effects for both devices on the 5th gen iPod Touch performance profile potentially leaving some performance unutilized in the iPhone 4S (although I suppose it would maintain a more stable frame rate)?

 

[tangey]

http://www.macotakara.jp/blog/ipod/entry-17953.html

I wonder if that means game developers will optimize separately for the 5th gen iPod Touch and iPhone 4S (not the just resolution difference but simplify effects for the iPod Touch) or just base the level of graphics effects for both devices on the 5th gen iPod Touch performance profile potentially leaving some performance unutilized in the iPhone 4S (although I suppose it would maintain a more stable frame rate)?

Or just letterbox the games on the touch ?

 

[ltcommander.data]

http://cens.com/cens/html/en/news/news_inner_41728.html

Apple's reportedly verified TSMC's 20nm process this August with plans for risk production in November and aiming to be in full production in Q4 2013 if all goes well. They're thinking this will be a quad core chip for the 2014 iPad with the 2014 iPhone remaining dual core. Interestingly, they believe Samsung has higher production costs than TSMC due to "the company's inefficiency".

http://blogs.wsj.com/digits/2012/10/11/chip-design-luminary-leaves-samsung-for-apple/

As well, Apple has picked up Jim Mergard, a former AMD vice-president and chief engineer, who worked on Brazos.

 

[anexanhume]

If that's true, I wonder if they'll bother with 28nm from either fab.

 

[ltcommander.data]

If that's true, I wonder if they'll bother with 28nm from either fab.

Well Samsung says shrinking from their 32nm process to their 28nm process is a very straight forward process. At the very least, it'd make sense for Apple to do a direct shrink of the A6 to Samsung's 28nm process for use in next year's iPad Mini, iPod Touch, and possibly an Apple TV refresh. I'd imagine Apple will want to stick with Samsung's proven 32nm process for the 2014 iPad's SoC. TSMC's 20nm process wouldn't be ready for next year's iPhone so that seems like a good candidate for an all new SoC on Samsung 28nm, which may be the last Samsung fabbed Apple SoC.

 

[anexanhume]

Anandtech published their iPhone 5 review. Great look into the A6, or 'Swift'.

http://www.anandtech.com/show/6330/the-iphone-5-review

 

[ltcommander.data]

http://www.anandtech.com/show/6330/the-iphone-5-review

Anand's iPhone 5 review is up.

Apparently the new A6 CPU is called Swift. It's more closely related to the Cortex A9 than A15. The number of execution units doesn't appear to have increased, instead the emphasis is on better utilization through a wider 3 decoder front-end and moving from 3 to 5 execution ports. Pipeline depth has increased being between the Cortex A9 and A8 and similar to Krait. Cache and memory latency has decreased and of course memory bandwidth has increased. Final conclusion is that Swift seems to be more Apple's answer to Krait than Cortex A15 and the end result is slightly ahead of Krait in performance and power profile.

EDIT: Oops late to post when I stepped away before finishing.

 

[wco81]

Yeah I wonder if iPhone 6 will get a whole new CPU core and Rogue. Probably will since I just bought the iPhone 5.

Also wonder what's in store for iPad 4. March may be too early for new CPU core and Rogue even if they wanted to pick these up. Plus there are rumors of an updated iPad (new connector, maybe more efficient SOC to reduce costs) to accompany the iPad Mini (which is officially going to be unveiled next Tuesday, 10/23).

 

[Pressure]

Since they now have full control, I would be surprised if they didn't make changes for the next iPhone iteration.

Especially if they start to use 28 nm or lower.

The iPad would certainly see its own SoC based on the Swift design, perhaps with bigger memory bandwidth due to the large resolution it has to drive.

Wouldn't be that surprised to see the new iPad Mini / Nano / Air with the 32nm A5r2 though, although that is a bit like a Tegra 3 device these days, on its last leg.

 

[Ailuros]

Since they now have full control, I would be surprised if they didn't make changes for the next iPhone iteration.

Especially if they start to use 28 nm or lower.

The iPad would certainly see its own SoC based on the Swift design, perhaps with bigger memory bandwidth due to the large resolution it has to drive.

While I'd wish that the next iPad will be a Swift + Rogue combination under 28nm I'm not that sure it'll be the case after all.

Wouldn't be that surprised to see the new iPad Mini / Nano / Air with the 32nm A5r2 though, although that is a bit like a Tegra 3 device these days, on its last leg.

On the other hand how many T3 based devices are out there with a 1024*768 or equivalent resolution? Besides since A5R2 is on 32nm there's nothing that speaks against higher than 250MHz frequencies for the GPU without affecting too much if any power consumption. A mere 100MHz increase in GPU frequency would give it a rough estimated leap of 50% in stuff like GL_Benchmark2.5 over a T3 tablet.

 

[tangey]

I emailed anand to point out that the gpu table he has been using for a while and included in both the iphone5 and a6 pieces, was flawed in relation to the "as shipped" gpu speed for the ipad2 and 3. His table showed both running at 200mhz, when in fact they both run at 250mhz. This metric for ipad2/3 graphics was in part used by him to estimate the clock for the 543mp3 in the iphone5 as being 266mhz.

I suggested to him that the relative gpu performance between iphone4s/ipad3/iphone5 ( fill rate in particular, but many others too), led me to believe that the 543mp3 in the iphone5 was running at greater than 300mhz and my guess was 325mhz. The later revelation that the CPU was running at 1.3ghz led even more credence to it being 325mhz (1.3ghz/4).

He was good enough to response and says he's reviewed the data and agrees its most likely running somewhere north of 300mhz, and has removed the "as shipped" row from the gpu performance table.

 

[Pressure]

While I'd wish that the next iPad will be a Swift + Rogue combination under 28nm I'm not that sure it'll be the case after all.

On the other hand how many T3 based devices are out there with a 1024*768 or equivalent resolution? Besides since A5R2 is on 32nm there's nothing that speaks against higher than 250MHz frequencies for the GPU without affecting too much if any power consumption. A mere 100MHz increase in GPU frequency would give it a rough estimated leap of 50% in stuff like GL_Benchmark2.5 over a T3 tablet.

Nothing keeps Apple from clocking it higher than 250 Mhz but why not just retain the same performance target as the iPad 2? Much easier for developers and they do not have to scale graphic performance with their 3D games.

Of course, they could throw a curve ball and introduce the iPad Mini with A6 or a 32 nm shrink of the A5X.

We'll know soon enough

He was good enough to response and says he's reviewed the data and agrees its most likely running somewhere north of 300mhz, and has removed the "as shipped" row from the gpu performance table.

Good to hear he listens to reason heh.