Next-Gen iPhone & iPhone Nano Speculation

[Ailuros]

Apple doesn't obviously concentrate mostly on CPU processing especially since A5. When it comes to die area the CPU and GPU blocks capture give or take the same sqmm from the entire die estate. Since the A5 successor will be most likely manufactured under a smaller process than 45nm/A5, it'll be interesting to see how and where they'll invest the additional area resources they'll gain from the smaller process.

When it comes to 3D I'd rather have something like in the PS Vita SoC (4*A9+SGX543MP4) then just a quad A9 or whatever alternative configuration with a times weaker GPU block. And before comes along with the same tired line that they hypothetically "don't need" as much additional GPU power, if they're really aiming for a resolution like 2048*1536 or similar, the need for a lot more fill-rate is definitely there amongst others.

Tegra3's GPU might be relatively strong for today's comparisons, but at 1080p already its barely on Tegra2 GPU level of performance. That's a very strong CPU, paired with a rather mediocre GPU; unless of course the device has a 720p only screen.

 

[Entropy]

Apple doesn't obviously concentrate mostly on CPU processing especially since A5. When it comes to die area the CPU and GPU blocks capture give or take the same sqmm from the entire die estate. Since the A5 successor will be most likely manufactured under a smaller process than 45nm/A5, it'll be interesting to see how and where they'll invest the additional area resources they'll gain from the smaller process.

When it comes to 3D I'd rather have something like in the PS Vita SoC (4*A9+SGX543MP4) then just a quad A9 or whatever alternative configuration with a times weaker GPU block. And before comes along with the same tired line that they hypothetically "don't need" as much additional GPU power, if they're really aiming for a resolution like 2048*1536 or similar, the need for a lot more fill-rate is definitely there amongst others.

I agree that their next-gen SoC will probably be very similar but merely increasing the number of cores is much easier than changing the CPU or GPU was in the old days. So given their area budget I don't think a straight shrink is very likely either.

So if the new SoC is based on finer lithography, it would seem that we have a consensus that increasing the number of SGX543 blocks would take priority over adding CPUs, but that adding A9 CPU cores would be relatively simple from a technological standpoint. I think dagamer had a valid point in that if this would occur, this would almost imply that Apple would offer quad A15 cores as well. It is difficult to see them retreat in the spec numbers game.

When the specs for the A5 processor started to surface, and on the same process as the A4 even, I assumed that the large increases in processing power implied that the higher resolution displays would be introduced at the same time. It didn't occur. I still feel that higher resolution displays implies higher processing capabilities to maintain overall responsiveness, so unless there was a syncing glitch with the last model, I think it is safe to assume that they will provide a beefier SoC with the introduction of the new screens.

Ailuros' 4*A9+SGX543MP4 suggestion seems both reasonable and straightforward. (Even though I agree with dagamer that the actual need for quad-CPUs is limited.) But that leaves us with two questions.
1. What other modifications of the system would be desirable in order to benefit optimally from the additional ALU resources?
2. Are there reasonably realistic alternatives in terms of CPU and GPU cores with which Apple could surprise us and the competition? It is difficult for me to judge the immediate viability of the A15 for instance, as well as the various SGX core and configuration alternatives.

 

[Arun]

So if the new SoC is based on finer lithography, it would seem that we have a consensus that increasing the number of SGX543 blocks would take priority over adding CPUs, but that adding A9 CPU cores would be relatively simple from a technological standpoint. I think dagamer had a valid point in that if this would occur, this would almost imply that Apple would offer quad A15 cores as well. It is difficult to see them retreat in the spec numbers game.

Agreed, although as I said I see a very small chance that they'd do something completely crazy (in a good way) like 2xA15+4xA7 with both being usable at the same time.

Just to be clear though: I think the most likely path for Apple is sticking to 2xA9 on 32nm (assuming they're still at Samsung) and moving to 2xA15+2xA7 on 28nm (either at Samsung or TSMC). I just wanted to highlight that other options can make very good technical sense and not necessarily be a gimmick in the way that dagamer implied.

2. Are there reasonably realistic alternatives in terms of CPU and GPU cores with which Apple could surprise us and the competition? It is difficult for me to judge the immediate viability of the A15 for instance, as well as the various SGX core and configuration alternatives.

The A15 would not be available in the iPad 3 timeframe. SGX544 and SGX554 would (probably) be available but obviously slightly more work for Apple to integrate - honestly not sure how much.

 

[Ailuros]

Entropy,

My point was rather that it would make more sense for Apple's so far strategy to have a finer balance between CPU and GPU processing power, or else a config that wouldn't leave the system either CPU or GPU bound. It goes without saying that the newer the architecture, the better it would be for both sides but availability of specific IP unfortunately plays a role too as Arun said.

And no I don't agree that a quad A9 would be somewhat redundant or "overkill" in a case scenario where it would be theoretically too early for anything A15 and you wouldn't want to increase in a dual A9 case the living shit out of its frequency to reach performance target N. Same goes for the GPU block too. There's a quite a bit of pressure from all the competing sides and I don't think Apple has the luxury at this stage to lean back and go just for a quick direct shrink and call it a day; definitely not if they're aiming for a super high resolution.

Arun,

I'm not familiar with Apple's plans, but given their iOS the reasons why they'd want something like a 544 or 554 don't seem all that clear to me. If they're going for a higher GPU core count one of the good reasons would be N times more fill-rate. 554MPx would give a crapload of more FLOPs, but not more fill-rate. Neither 544 nor 554 can be smaller in die area than a 543; I'd personally invest that die area difference in higher frequencies, but that's just me.

 

[dagamer]

The problem with your analysis is that it makes sense.
A bit more seriously, and this one I mean, is that the scheme is too predictable.
Regrettable as it may be, part of the design of any consumer device is marketability. And if Apple believes that superficial tech websites are in any way a significant part of what shapes market perception of their devices, then that might be reflected in SoC designs. GIven that multi-threaded cache resident benchmarks are the order of the day at such sites, and the lack of significant (for Apple) die cost of adding cores, particularly at 32/28nm, going quad A9 could simply be an issue of playing it safe, marketing wise. Quad has to be better than dual, right? Going to quad A15 would be the next probable step. It would cost a bit, but not too much, and Apple is in a good position to put the squeeze on their competition in that regard.

The big question is how aggressive Apple wants to be. The tablet market is theirs to loose, and they have been aware for some time that Android will be joined by a Wintel attack, come Windows 8. Increasing screen resolution and quality is but one way to stay a step ahead of the competition, they are bound to use any other hardware means at their disposal as well, and SoC cost/complexity is one area where they are capable of being very competitive. Caches, data paths, GPU width, core count et cetera are all examples of where they could strengthen their offering and distance themselves from the also-rans. If the cost in terms of battery life is small enough, I wouldn't be surprised to see Apple going above and beyond the scheme you outlined, for market positioning reasons rather than purely technical ones, basically, ensuring that technological predictability isn't taken advantage of. I'd say they already did this with their A4->A5 move on the same lithographic process.

PS. Unfortunately the Cortex A9 doesn't really support going beyond four cores. Otherwise it would have been wonderfully ironic to see Apple put 16 A9s on their SoC, typically "turned off" obviously, but there when called upon to play the numbers game.

If Apple falls for the trap of sticking in 4 A9 cores @ same clock speed as the current A5 and thinking that's "advancement" in the spec race, their hardware design team needs a cold slap to the face. I think everyone on this forum has a firm understanding that more cores doesn't not automatically equal better performance, and the law of diminishing returns definitely comes into play. I'd wager that by far most of the speed increases that come from real-world apps (not benchmarks) have to do with the A5 having a Cortex A9 CPU than being dual-core. I would bet that 99% of app developers never write multi-threaded code (I certainly never have) and probably never will. And that is why a quad-core CPU is a horrible idea on a mobile platform.

But if Apple wants to play the game than 4 > 2 and continue with that line of thinking in the future, they are dooooooomed.

 

[dagamer]

Agreed, although as I said I see a very small chance that they'd do something completely crazy (in a good way) like 2xA15+4xA7 with both being usable at the same time.

Just to be clear though: I think the most likely path for Apple is sticking to 2xA9 on 32nm (assuming they're still at Samsung) and moving to 2xA15+2xA7 on 28nm (either at Samsung or TSMC). I just wanted to highlight that other options can make very good technical sense and not necessarily be a gimmick in the way that dagamer implied.

Not so much a gimmick in that a quad core CPU only benefits apps that are written to be heavily multi-threaded (and how many of those really are there on the iPhone?), as compared to a 50% clock speed increase which every single of the 500,000 apps on the App Store would benefit from. From a user experience and even developer stand point, I think the decision is quite easy.

The A15 would not be available in the iPad 3 timeframe. SGX544 and SGX554 would (probably) be available but obviously slightly more work for Apple to integrate - honestly not sure how much.

As far as I can remember, the SGX 544 only adds stuff related to DirectX 9 compliance (likely for Windows 8), and the SGX 554 was announced only 2 months before the SGX 6 series (Rogue), so I doubt that's really ready for use. I feel the only reason Apple went with such a powerful GPU for the A5 is because they knew they wanted to release an iPad 3 with a high res display and needed the power to feed that many pixels, plus save on development costs by designing the chip once and having a minor revision rely on die shrinks for a CPU clock speed bump without increasing power consumption.

 

[wco81]

When they introduced the A5 SOC at the iPad 2 announcement, they said it was dual core but didn't mention that it was A9 instead of A8 nor did they mention the clock speed or RAM.

They didn't specify the GPU model either, just that it was up to 9 times faster.

Not that any of that hints at which direction they might go for their next SOC.

 

[dagamer]

My magic 8 ball says the Apple A6 we get this year is a 1.5Ghz dual-A9 with an unchanged PowerVR SGX543MP2. Then in 2013, we'll see an Apple A7 with a dual-A7/15 in a big.LITTLE arrangement and a PowerVR 6 series MP core. Not too exciting, very predictable, but people don't' buy iPhones for hardware but software!

 

[silent_guy]

I would bet that 99% of app developers never write multi-threaded code (I certainly never have) and probably never will. And that is why a quad-core CPU is a horrible idea on a mobile platform.

With the availability of NSOperation, writing programs where side threads are executing lower priority task is dead easy (e.g. loading and saving data in the background) and it makes a big difference in the responsiveness of an app. Another classic example are gaming engines that decouple the render engine (always running at 60Hz) and object logic. ID Software is doing this for one of its titles (forgot which, but the source code is available) and it's really neat.

Not saying that this will make a noticeable difference when going from 2 to 4 cores, but 1% is a very low barrier to cross when making wagers.

 

[ltcommander.data]

My magic 8 ball says the Apple A6 we get this year is a 1.5Ghz dual-A9 with an unchanged PowerVR SGX543MP2. Then in 2013, we'll see an Apple A7 with a dual-A7/15 in a big.LITTLE arrangement and a PowerVR 6 series MP core. Not too exciting, very predictable, but people don't' buy iPhones for hardware but software!

If we are taking bets on what we think the Apple A6 will look like, I'll stick with 2 x 1.6GHz dual-A9 + SGX543MP4 for iPad 3 and 2 x 1.2GHz dual-A9 + SGX543MP4 for next-gen iPhone. The iPad got a 66% clock speed increase in it's Cortex A8 over the iPhone 3GS so 50-60% clock speed increase between the A5 and A6 is reasonable. Going MP4 on the GPU makes sense assuming Apple is more conscious of the need for more fill-rate when going to a Retina iPad as Ailuros stresses. There may be a resolution change/increase for the next-gen iPhone too if the much rumoured 4" screen pans out while maintaining the Retina moniker so MP4 may not be as overkill as it seems there. OpenCL support for the SGX543 will probably also be enabled with iOS 6 so the GPGPU can help mitigate concerns that GPU performance is increasing disproportionality faster than CPU.

Ailuros' 4*A9+SGX543MP4 suggestion seems both reasonable and straightforward. (Even though I agree with dagamer that the actual need for quad-CPUs is limited.) But that leaves us with two questions.
1. What other modifications of the system would be desirable in order to benefit optimally from the additional ALU resources?
2. Are there reasonably realistic alternatives in terms of CPU and GPU cores with which Apple could surprise us and the competition? It is difficult for me to judge the immediate viability of the A15 for instance, as well as the various SGX core and configuration alternatives.

Going from the iPhone 3GS to the iPhone A4, the Cortex A8 had it's L2 cache increased from 256KB to 512KB. The Apple A5 implements a shared 1MB L2 cache, so per core cache is unchanged at 512KB per core. Would it be worthwhile to double the L2 cache in the Apple A6 to a shared 2MB or 1MB per core?

Would it be a significant amount of design work to give the GPU direct access to this enlarged shared L2 cache similar to how Sandy Bridge/Ivy Bridge's shared L3 cache is accessible to both the CPU cores and GPU? Presumably, this would be beneficial for OpenCL in terms of latency and bandwidth and perhaps help in gaming performance too.

As another thought, if mobile GPUs tend to be memory bandwidth limited, could Apple elect to squeeze more mileage from a SGX543MP2 configuration by doing what Sony did and modify it to add dedicated VRAM?

 

[dagamer]

With the availability of NSOperation, writing programs where side threads are executing lower priority task is dead easy (e.g. loading and saving data in the background) and it makes a big difference in the responsiveness of an app. Another classic example are gaming engines that decouple the render engine (always running at 60Hz) and object logic. ID Software is doing this for one of its titles (forgot which, but the source code is available) and it's really neat.

Not saying that this will make a noticeable difference when going from 2 to 4 cores, but 1% is a very low barrier to cross when making wagers.

Besides your top tier games, can most developers really afford the time to optimize for high-end hardware? Then the best selling game on iOS is Angry Birds that can technically run on even the original iPhone, will they put in that energy?

I mean, you have to keep in mind that even the best of developers has to realize that the number of multi-core iOS devices out there isn't a majority of iOS devices out there, and because the iPod Touch is still based on the A4, single-core performance tweaking is still where you get the most bang for your buck.

By the way, does anyone have stats on how multi-thread code performs on a single core CPU? I would assume there is some overhead from having competing threads, but it's not terrible.

 

[tangey]

I'll be surprised if Apple produce an ipad3 with 4 times the pixels, and don't uplift the graphics.

Ok, there is significant untapped graphics performance in the existing A5, which is revealed in the frame-rate unlocked figures at glbenchmark. Those figures suggest that the screen res could be increased x1.5-x2.5, and the benchmarks would still hit max frames in games/apps (assuming the benchmark is representative of general preformance). So adding two cores and not doing anything with the clock, would give enough uplift to provide the similar graphics performance on a higher res screen. That scenario does provide some options for Apple in regard to running a "power saving mode" where 2 of 4 cores are turned off when running in 1028x768 mode ? Or go from 250Mhz to 400Mhz on 2 cores.

Yes, the alternative is they leave the graphics unchanged, can market it as being able to play existing games at exactly the same speed (in 1028x768 mode) and also capability of support hi res apps (at 1/3-1/2 the performance). However given that they have used graphics performance as a very prominent marketing tool on both the ipad2 and iphone4S, I'm not sure thats the direction they'll take.

 

[wco81]

Would developers targeting iPad games (HD) make the same choice as console games developers to use lower resolution with more effects?

Especially for this year, when there are tens of millions of iPads already running at the lower resolution?

It will take awhile for a higher-res iPad 3 to build up a sizable installed base.

 

[silent_guy]

Besides your top tier games, can most developers really afford the time to optimize for high-end hardware? Then the best selling game on iOS is Angry Birds that can technically run on even the original iPhone, will they put in that energy?

I'm talking 3GS and 4 here, no high-end HW at all and definitely not dual core. The multi-threading makes a huge difference here wrt to fluidity and stuttering. If you put texture loading or other IO background tasks in the same message loop, it can really mess things up. A big no-no for a device that got its fame for being smooth.
A lower priority thread solves that very nicely. And it's so easy it'd be stupid not to do it.

Multi-core only makes this better.

I think you'll have a hard time finding any app that isn't at some point during its execution lifetime multi-threaded.

By the way, does anyone have stats on how multi-thread code performs on a single core CPU? I would assume there is some overhead from having competing threads, but it's not terrible.

No stats, but you don't need those to see that absolute performance will usually go down a little while perceived performance goes way, way up.

 

[Entropy]

No stats, but you don't need those to see that absolute performance will usually go down a little while perceived performance goes way, way up.

As you're obviously are aware, but don't want to spell out, whether perceived performance improves or not depends on exactly what you are trying to achieve with your multi-threading, and how the OS environment handles that. The underlying computing reality is, as was already indicated, that the absolute performance typically drops a bit.

Problems with I/O handling, performance issues involving system software "layering", et cetera, are best handled where they are created - in software. The software stack is way too neglected at technology enthusiast sites, probably because it requires a lot more understanding, and is difficult to evaluate, much less compare. But from an importance point of view it really should get a lot more attention, and for that matter, credit where due.

PS. For better or worse, Apple says: "Do it our way, or don't bother trying to peddle your stuff to our customers." That doesn't only cover texture compression . It has been their way since they, together with the launch of the first mac in 1984, also published their rather substantial and well thought out user interface guidelines, and basically said "do it like this, or you'll mess up the user experience consistency and idea of the entire platform". That level of insight was what initially impressed me more even than the WIMP interface itself, back in the day. (In the context of the issues discussed, human interface responsiveness was a key precept, integral to how the interface was designed. It's fascinating to see that it's just as relevant 30 years down the line, despite the enormous advances in computing power since then. Hardware multithreading is really not that important, it's all about the software.)

 

[Ailuros]

My magic 8 ball says the Apple A6 we get this year is a 1.5Ghz dual-A9 with an unchanged PowerVR SGX543MP2. Then in 2013, we'll see an Apple A7 with a dual-A7/15 in a big.LITTLE arrangement and a PowerVR 6 series MP core. Not too exciting, very predictable, but people don't' buy iPhones for hardware but software!

Assuming they'd go for such a scenario, why would they increase the CPU block frequency by 50% and leave the GPU block completely unchanged?

Up until A5 appeared nobody wanted to believe that Apple would go for a SGX543MP2 in the first place because they supposedly were too "conservative" with hw. A5 proved them wrong and if one thing Apple didn't care in the least about die area, simply because with the volumes they're dealing it's not a cost issue for them.

Your scenario suggests that they suddenly care about die area and are going for a die shrink with a higher CPU frequency only. The next iPhone might not appear before fall this year and even if they'd chose to increase the resolution I severely doubt that they'd go beyond 1024, in a less likely case even up to 720p. Likeliest scenario IMHO being that they might stick to 960*540. That aside if their next iPad truly goes up to 2048*1536, which is quite a brutal increase in resolution with the exact same GPU block it sounds more like they'll end up with less performance than on iPad2. Let's be generous and say that they might break even. The problem then will be that they won't have a single advantage against any other tablet SoC.

My crystal ball says that for example OMAP4470 would break even with the hw scenario you're suggesting and OMAP5 beating it by more than a factor 2x. At least when it comes to graphics Apple A5 is even to date the fastest solution you can get. I'm sure Apple is willing to kill that advantage because it suddenly worries about manufacturing costs.

 

[Pressure]

Just a small correction but the iPhone has a 640 x 960 resolution ;-)

I fully agree. What we have seen so far from Apple is that they are very aggressive in the SoC they choose and I wouldn't be surprised if it turned out that the A6 would be the fastest SoC of 2012.

 

[Helmore]

Just a small correction but the iPhone has a 640 x 960 resolution ;-)

I fully agree. What we have seen so far from Apple is that they are very aggressive in the SoC they choose and I wouldn't be surprised if it turned out that the A6 would be the fastest SoC of 2012.

The A6 the fastest? I can see that happening with regards to graphics performance, but not really in CPU performance. My guess would be something like a much higher clocked dual core Cortex-A9 and a higher clocked SGX543 MP2 or MP4 for the Apple A6. A well, we'll probably get to see what the A6 is like this quarter anyway.

 

[dagamer]

I'm talking 3GS and 4 here, no high-end HW at all and definitely not dual core. The multi-threading makes a huge difference here wrt to fluidity and stuttering. If you put texture loading or other IO background tasks in the same message loop, it can really mess things up. A big no-no for a device that got its fame for being smooth.
A lower priority thread solves that very nicely. And it's so easy it'd be stupid not to do it.

Multi-core only makes this better.

I think you'll have a hard time finding any app that isn't at some point during its execution lifetime multi-threaded.


No stats, but you don't need those to see that absolute performance will usually go down a little while perceived performance goes way, way up.

I am separating "apps" from "games" which rarely need to optimize anything but the worse performing lines of code since nothing runs in a time-sensitive 30/60fps loop but is event driven. I am also not arguing that apps are never multi-threaded, but most apps do not have a lot of incentive to write a ton of multi-threaded code whereas a game would like to squeeze out as much performance as possible.

 

[dagamer]

Assuming they'd go for such a scenario, why would they increase the CPU block frequency by 50% and leave the GPU block completely unchanged?

Up until A5 appeared nobody wanted to believe that Apple would go for a SGX543MP2 in the first place because they supposedly were too "conservative" with hw. A5 proved them wrong and if one thing Apple didn't care in the least about die area, simply because with the volumes they're dealing it's not a cost issue for them.

Your scenario suggests that they suddenly care about die area and are going for a die shrink with a higher CPU frequency only. The next iPhone might not appear before fall this year and even if they'd chose to increase the resolution I severely doubt that they'd go beyond 1024, in a less likely case even up to 720p. Likeliest scenario IMHO being that they might stick to 960*540. That aside if their next iPad truly goes up to 2048*1536, which is quite a brutal increase in resolution with the exact same GPU block it sounds more like they'll end up with less performance than on iPad2. Let's be generous and say that they might break even. The problem then will be that they won't have a single advantage against any other tablet SoC.

My crystal ball says that for example OMAP4470 would break even with the hw scenario you're suggesting and OMAP5 beating it by more than a factor 2x. At least when it comes to graphics Apple A5 is even to date the fastest solution you can get. I'm sure Apple is willing to kill that advantage because it suddenly worries about manufacturing costs.

I feel a bit iffy on the GPU scenario, but unless Apple changed the clock speed of the SGX 535 in the A4 compared to whatever is in the 3GS, it means they designed a chip to last two generations without anything more than CPU clock speed increases from die shrinks. Then again, none of the previous CPU/GPU designs could be used in a multi-core scenario, I could be totally wrong.

I don't really predict the iPhone screen to change radically unless they just decide "screw it, we're going with a 4.3" screen and 720p display". Highly unlikely, but you never know.