Next-Gen iPhone & iPhone Nano Speculation

[juicytuna]

The A5 on 45nm already has a die size of 120mm2. A quad A9 with a possible MP4 on the same process (AFAIK Samsung don't have 40nm?) would be humongous for a SOC.

We know demand for i-devices is increasing at a phenomenal rate and I doubt apple would want to be in a position where they might be supply constrained by the SOC.

I think what we will see is A5 with higher clocks and possibly a die shrink to 32nm.

 

[wco81]

Read somewhere that sales of iPad in the fourth quarter of last year was constrained not by Kindle Fire or any other tablet but by iPhone 4S. Many of the people who bought the iPhone could only buy one at that time and chose the iPhone over the iPad.

But the product is showing its age, so they will have to come out with the new model, whether or not it's the best timing, regardless of the lithography.

 

[Arun]

That's some very good analysis indeed! I don't think the heatspreader means that much though - if you can afford it, it's certainly a win even at the A5's power level (lower temp -> lower leakage and power in general).

 

[benjiro]

The A5 on 45nm already has a die size of 120mm2. A quad A9 with a possible MP4 on the same process (AFAIK Samsung don't have 40nm?) would be humongous for a SOC.

*Oeps* You are right, made a mistake on that one.

Samsung's PS VITA SOC is 45nm.

But, coming back to this. The Vita SOC is already a Quad Core CPU + SGX543MP4+ @ 45nm. In other words, size does not exactly mater.

One of the arguments that i hear in the past, is that the 120mm² core of the A5, is possible because Apple is not into reselling there SOC anyway. They are not restrained intro maximum yield like those manufactures that resell there SOC. The same can be said about Sony with the PS Vita.

Read somewhere that sales of iPad in the fourth quarter of last year was constrained not by Kindle Fire or any other tablet but by iPhone 4S. Many of the people who bought the iPhone could only buy one at that time and chose the iPhone over the iPad.

Was it not more the fact, that people stopped buying the iPhone4, in favor of waiting for the iPhone4S ( or iPhone5 as most expected it to be ). The Q3 numbers showed a massive drop in iPhone4 sales, what was recovered by massive sales of the 4S in Q4.

That's some very good analysis indeed! I don't think the heatspreader means that much though - if you can afford it, it's certainly a win even at the A5's power level (lower temp -> lower leakage and power in general).

I think some parts of the SOC get too hot, and by using a heat spreader this can be more spread out. Heat spreaders are very normal in the PC world for how many years?

One of the big reasons that the PS Vita is this "old" design, was that they had problems with the heat from there SOC.

Seeing that Heat spreader on the SOC, really makes me wonder if its going to be 32nm... Looks more like 45nm then? Unless this was a prototype.

32nm?

If its 32nm, technically they need have a much lower heat problem, so there is no reason for a heat spreader on it. Unless! They also increased the CPU/GPU clock ( but then its not just heat, but also power consumption ).

If they stick to 45nm, that is like 3 generations already. Apple A4, A5, and now A5X ... Add a maybe more power hungry screen. Even with bigger battery size as we seen on those leaked pictures, ... I don't know. Looks odd to still stick to 45nm.

The data is very conflicting on this.

Samsung is was planning on sampeling there Exynos 4212 in Q4. This is made on 32nm. So there is already a move to 32nm by Samsung.

This fits with the November Production date of that Apple A5x that we see. And then we come back to that blasted heat spreader.


Its going to be interesting, to see what they really did end up with. There are plenty of scenario's that can be played out.

* Cortex A9 looks to be almost 99% chance.
* Quad Core ... that is more like 70% chance.
* Manufacturing process ... That is a massive unknown, and a big factor in the Power usage, CPU / GPU Frequency, Die size ( less important ), etc...

Knowing?

17 days to go, before we know 100%.

Unless somebody buys that Apple A5x sample, and puts it under a microscope. Its not the first time that a few hardware sites have done this. Doing so, you instantly know the amount of CPU / GPU configuration, and you can also know the manufacturing process.

Interesting times ahead

 

[Rys]

But, coming back to this. The Vita SOC is already a Quad Core CPU + SGX543MP4+ @ 45nm. In other words, size does not exactly mater.

It does.

With a stacked DRAM package you can't grow the SoC die bigger than it. Vita has the DRAM external, so the die can be bigger.

 

[Lazy8s]

Again, using a 543MP2 after a 535 doesn't signify any change of strategy at Apple. They update their processor cores every two years... they pick as aggressive a core as is reasonable at that time... the 535 was just as aggressive a choice for its time (other phones were going for half the TMUs with the 530.) The 543MP2 was not overkill; it was the logical choice based upon die size and power considerations for its time for a very high end configuration, just as the 535 once was.

I don't see Apple bumping the CPU clock speed much, if at all. It's a huge battery hog and provides relatively small returns.

I hope it'll be 543MP4 + A9 quad at the same clocks on Samsung's 32nm LP HKMG. A7 should then be G6400 + custom Apple ARMv7 quad with a heterogeneous mix of cores.

 

[Lazy8s]

The Vita SoC is apparently being manufactured at the fabs they set up with their PS3 manufacturing partners, not Samsung.

 

[tangey]

Samsung's PS VITA SOC is 45nm.

But, coming back to this. The Vita SOC is already a Quad Core CPU + SGX543MP4+ @ 45nm. In other words, size does not exactly mater.

One of the arguments that i hear in the past, is that the 120mm² core of the A5, is possible because Apple is not into reselling there SOC anyway. They are not restrained intro maximum yield like those manufactures that resell there SOC. The same can be said about Sony with the PS Vita.

Vita packaging is much bigger than an iphone (working on the assumption that this chip will also be used in a phone), so it has much better chance at dissapating the heat. Also Vita I assume has a much bigger battery than a phone. Would the vita's Soc be viable in a phone size device (both from battery life and power dissapation ?)

Perhaps the bigger die size of the A5 was also to do with giving a bigger silicon area to get rid of heat. To go from A5 to a quad-cpu, quad-gpu using similar layout as A5 might mean the chip could be around 180mm2 if still on 45nm.

might be interesting to try to compare the picture with a picture of the ipad2 PCB to see if it could be determined if the physical packaging of the Soc has increased (could assume that the chip is the size of the raised part of the heatsink)

 

[wco81]

Vita has terrible battery life, doesn't it?

Battery life is one of the key things Apple touts, especially for iPads.

So unless yields at 32nm are really poor, you would think they'd have every motivation to try to make the smaller process work.

 

[juicytuna]

Vita gets approximately 3 hrs battery life in games with a 2210 mah battery and the Ipad2 gets 5 to 6 hours in 3d intensive games with ~6500 mah battery, so I would say they offer comparable battery life with battery size taken into account. It's the screen that accounts for the majority of the power drain anyway.

 

[Pressure]

Apple can afford the die size of the current A5 because of quantity.

When you cover the cost over 37 million iPhones and 18 million iPads, the cost differential from a 120nm to a 80nm product shouldn't be your main concern but whether or not you have enough parts.

I would be surprised if the A6 isn't made on 32 nm though, which seems to be the reason for the heat spreader. Less area to dissipate the heat over.

Well, that is my opinion at least

 

[Deleted member 13524]

I think the new chip is just an up-clocked A5 made on a different process.
Otherwise, it'd be called A6 and not A5X.

1.5GHz CPU, 300MHz GPU and dual-channel LPDDR-800 (basically, everything * 1.5) would be a good spec bump. Good enough for the increased resolution, at least.

- Qualcomm has been showing Snapdragon S3 chips running @ 1.5GHz with no power consumption deficiencies over dual CorteA9 1GHz solutions.
- nVidia has a quad-core because it needs to play a specs game for marketing reasons, which Apple doesn't need to (in fact, the company usually runs away from spec comparisons).
- PSVita has a quad-core because low-level optimizations can take good advantage of all the cores for games. The same cannot be said of a smartphone platform with lots of "amateur" development in mind.

 

[tangey]

I think the new chip is just an up-clocked A5 made on a different process.
Otherwise, it'd be called A6 and not A5X.

1.5GHz CPU, 300MHz GPU and dual-channel LPDDR-800 (basically, everything * 1.5) would be a good spec bump. Good enough for the increased resolution, at least.
.

GPU in ipad2 is clocking @250Mhz

 

[Mariner]

I think the new chip is just an up-clocked A5 made on a different process.
Otherwise, it'd be called A6 and not A5X.

1.5GHz CPU, 300MHz GPU and dual-channel LPDDR-800 (basically, everything * 1.5) would be a good spec bump. Good enough for the increased resolution, at least.

Sounds about right to me.

I'm not convinced that too much software is around which will be designed to use 4 cores and a clock speed boost allied with a smaller process could provide a very good amount of performance improvement as well as lower power usage. The 543MP2 is already an extremely capable GPU and a clock speed bump of 50% would still provide compelling performance.

Apple has avoided the mobile performance 'arms race' by keeping the A5 in the iPhone 4S at 800MHz when competing Android devices are on 1.5GHz. I don't see any reason why they wouldn't do the same with the iPad 3. A 'tick-tock' advancement akin to that of Intel's CPUs, perhaps?

 

[Deleted member 13524]

GPU in ipad2 is clocking @250Mhz

375Mhz GPU, then.

 

[wco81]

Why would they call it A5x? Even if they just make the slightest incremental changes, for branding reasons alone, why would they call it anything but A6? Not as if most people would look at electron microscope shots of the SOC and claim, "hey, it's the same cores."

It's one thing to call it the iPhone 4S because of the same form factor was retained. But for an SOC, there is no direct-mapping to what kind of cores or process is used.

 

[Ailuros]

375Mhz GPU, then.

Assuming Samsung 32nm just a 50% increase in frequency compared to 40nm should be doable. However IHVs don't always necessarily invest the entire "headroom" the smaller process allows just in frequency increases. In fact in the GPU world direct shrinks have become increasingly rare and the weight goes mostly to invest the headroom in a combination of more units with modest frequency increases.

 

[Deleted member 13524]

Assuming Samsung 32nm just a 50% increase in frequency compared to 40nm should be doable. However IHVs don't always necessarily invest the entire "headroom" the smaller process allows just in frequency increases. In fact in the GPU world direct shrinks have become increasingly rare and the weight goes mostly to invest the headroom in a combination of more units with modest frequency increases.

But the needed development effort between a direct shrink and actually changing functional units isn't the same, is it?
If the A5 is good enough with +50% clocks, why wouldn't the company prefer to focus its efforts in developing the A6 and A7 for a 2013 release?

 

[Lazy8s]

Whether the improvement is just speed increases or whether it's slotting in extra cores, this year's SoC could be A6 even if only by name.

The limiting factor on changing to new cores too frequently is probably the stability of the hardware and software platform rather than the amount of effort and resources Apple would have to spend. Changing to new cores too often would be too hard for the hardware team to implement well and too hard for the software developers (internal and external) to keep everything well refined.

 

[dagamer]

Sounds about right to me.

I'm not convinced that too much software is around which will be designed to use 4 cores and a clock speed boost allied with a smaller process could provide a very good amount of performance improvement as well as lower power usage. The 543MP2 is already an extremely capable GPU and a clock speed bump of 50% would still provide compelling performance.

Apple has avoided the mobile performance 'arms race' by keeping the A5 in the iPhone 4S at 800MHz when competing Android devices are on 1.5GHz. I don't see any reason why they wouldn't do the same with the iPad 3. A 'tick-tock' advancement akin to that of Intel's CPUs, perhaps?

I think everything is tick-tock when talking about Apple, with the phone design and hardware team on two-year overlapping schedules for major revisions, leaving in between years with minor updates (phone design) and die shrinks (SoC design).