Apple A9 SoC

Entropy

Veteran
It is time to start a thread for Apples new generation of SoCs, with the first informational leak potentially out.
Geekbench info.
If legitimate, 2.0 GHz constitutes an almost 50% increase in clock, promising quite remarkable single thread performance even if there would be no architectural enhancements.
 
Assuming another ~15% clock-for-clock performance increase from A8 to A9, we could see a Geekbench single-threaded score of ≥ 2500 !

The CPU details don't seem unreasonable, assuming it is still dual-core, but the 1 GB of RAM gives me some doubts about the legitimacy of this leak. Rumors have strongly pointed towards 2 GB of RAM for this year's iPhones.

By the way, I found a Geekbench result for an "iPhone8,2" (presumably an iPhone 6S) that features a 1.51 GHz triple-core processor. Could this result be genuine or is it just a renamed iPad Air 2? I have compared it to an iPad Air 2 result with similar overall scores and found negligible differences in most of the sub-benchmarks.
 
Last edited:
The A9 could be mainly a die shrink of the A8X, replacing dual memory bus with single LPDDR4.
That would still be the fastest tablet SoC around, and could thus serve both iPhone and iPad.
So no need for an A9X and similar to A7 used for both devices.
 
Last edited:
Assuming another ~15% clock-for-clock performance increase from A8 to A9, we could see a Geekbench single-threaded score of ≥ 2500 !
Even though TSMC claims 40%speed/60%power improvement for 16nmFF vs. 20SoC, I certainly didn't expect to see it materialise in the real world. Don't know how far they can push IPC at the same time. DDR4 may help a bit, or it may be required just to fully keep up with the improvements in CPU and GPU throughput. Density improvements are modest this time around, so I don't know what we can expect in terms of for instance L2/L3 cache sizes. A bigger, lower latency L3 (well, cache at all levels, actually) would be gravy of course, given that main memory latency will probably have increased a fair bit in terms of CPU cycles. I see them nipping and tucking with the Cyclone core, looking forward to seeing their progress.
I wonder at what generation Apple will drop 32-bit support, if ever. 64-bit is mandatory for software already so I guess it could conceivably happen at 10nm. I look forward to a new processor design from Apple, but it requires time, 64-bit app statistics for targeting, substantially changed lithographic possibilities and possibly something like dropping backwards compatibility cruft to make it really worthwhile.
Back to the present, if 2GHz pans out for the next iPhone, and it can be sustained for more than an AnTuTu run, it will be quite remarkable in its own right.
 
if 2GHz pans out for the next iPhone, and it can be sustained for more than an AnTuTu run, it will be quite remarkable in its own right.

That is quite unlikely. The increase from A7 to A8 was 100 Mhz,
Now an increase of 600 Mhz, right.
Even at 1.5Ghz, an A8 is faster than a 2Ghz A57.
And it can sustain that for longer without throttling.
 
Last edited:
Even though TSMC claims 40%speed/60%power improvement for 16nmFF vs. 20SoC, I certainly didn't expect to see it materialise in the real world. .

Yup, and it's usually an either/or proposition when it comes to power and performance. I really hope Apple focuses on lower power and more sustained performance again instead of some peak performance that's for benchmarks only. I think that's better for the end user.
 
9to5mac got a tip regarding the A9, although they say it may be fake.

An unsubstantiated tip suggests that the A9 inside the iPhone 6S achieved a Geekbench single-core score of 1921 points, and a multi-core score of 4873 points.
[…]
the same tipster claims that the A9 will feature two 1.7GHz cores and two 1.2GHz cores.
If true, then Apple may be using an approach resembling big.LITTLE. Does this seem plausible? I was under the impression that Apple was eschewing this type of configuration.

Also, I think the single-core score is odd. A hypothetical 1.7 GHz Typhoon with no additional changes should already achieve 1900-1950 points in Geekbench single-core, assuming linear or close-to-linear scaling. That's in contrast to the A7-A8 transition where some noticeable boosts were seen even after adjusting for clocks.
 
y
Yup, and it's usually an either/or proposition when it comes to power and performance. I really hope Apple focuses on lower power and more sustained performance again instead of some peak performance that's for benchmarks only. I think that's better for the end user.
Those percentages certainly constitutes either/or. There are a couple of reasons why Apple may actually choose to reap the benefits of a new process in the shape of increased clocks this time around.
FIrst, density doesn't improve all that much. A bit, certainly, but it may be that Apple prefers to keep the die size modest due to process uncertainties, and of course cost. If so, it may make sense to push clocks rather than making the design wider when shooting for increased performance. Optimisations in iOS helps with battery life.
Second, the new process actually allows significant improvements in clock for low power devices, more so than in the 28nm -> 20nm transition for instance. It's an opportunity both for power savings and for performance enhancements, but since this is a refresh of the old model, performance may be given priority, letting battery life remain similar. Apple needs to bring more to the table than force touch to keep consumer interest up for another year of outwardly identical phones. Performance is easy to enhance in this case.

The alleged Geekbench score is at odds with the screenshot that shows clocks and RAM. I tend to favour the screenshot a bit in terms of believability, both because I can't really make sense of the single thread score, don't believe Apple would go big.LITTLE, and because of the disappointing 1GB RAM of the screenshot. If it had been fabricated, I would have assumed it to go with the rumoured 2GB. We'll know shortly.
 
What I'd like is a SoC on Samsung's 14nm process with a PowerVR GT7800 @ 533 MHz paired to a tri-core (A8X-like set-up) Cyclone v3 (just incremental architectural enhancements) @ 1.6 GHz. 2 GB RAM, of course.

That should yield more than enough on the processing side for a S-year update to the iPhone line: iPhone 6 to 6s.

Yeah, that'd be a return to a wider, lower clocked design approach, but there's certainly headroom for the die area and an ever-present need for lower voltages.
 
My bet is on tri-core 1.85ghz with 2gb lpddr4.
Doesn't fit the single thread score very well though. (1.85/1.4)=1.32 multiply with A8 score yields 1.32*1622=2143.
The single thread score in the alleged leak would yield a 1.65GHz clock, all else being equal.
A tricore configuration like you propose is certainly possible. Pretty much the iPad Air2 A8x in CPU capabilities, bandwidth and RAM, so it's apparantly a balance Apple finds reasonable.
Since half the A8 is stuff other than CPU/GPU/cache/memory interface, and I honestly don't know how that "stuff" scales with the process shift, I find it difficult to ballpark die size changes for various hypothetical A9 configurations. But Apple deciding to add or cut back 15mm2 vs. the A8 would seem to make a huge difference in what they can squeeze into the A9 in terms if CPU/GPU.
 
And the silly season has truly started. :)
A9 and...A9x?!?
This is actually the A9 rumour that seems most reasonable so far. Don't know what to say about the A9x part here - what device is this number supposed to be measured from? Graphics score of the A9 in this rumour seems to fit a GT7400 at some reasonable clock.
 
I agree that it seems more reasonable than the other ones posted in this thread. Also, as far as I can tell, the "A9X" screenshot in the WCCFTech link does not specifically mention an A9X, just the iPad Air 3.

The same (similar?) source has claimed information for multiple upcoming SoCs (which honestly makes me more suspicious, since there are so many). Yesterday I made a spreadsheet containing the relevant details and benchmark scores.

https://docs.google.com/spreadsheet...okffHNxDXBPHaJX2Mg3rfUpcsog/edit#gid=52055047
 
Last edited:
With all the talk of the new A9 processor intended for the new iPhones, and the new A9x processor supposedly intended for the new iPad Air 3. What will the new iPad Pro get? If expected, the A9 to have 2 cores and A9x to have 3 cores similar to their previous generation counterparts. Will there be an additional SoC for the iPad Pro that will feature 4 cores, or will it be the A9x clocked at a higher speed and with more RAM.

Also with the release of the iPad Pro, are we likely going to see, true desktop/laptop performance from a fan-less tablet?
 
Well, there was a rumor that "iPad Pro" would not be shown at this upcoming event, but you know how these rumors are. Impossible to tell really.

Apple's event is in about a week, we'll know the details about the A9 then... :p
 
With all the talk of the new A9 processor intended for the new iPhones, and the new A9x processor supposedly intended for the new iPad Air 3. What will the new iPad Pro get? If expected, the A9 to have 2 cores and A9x to have 3 cores similar to their previous generation counterparts. Will there be an additional SoC for the iPad Pro that will feature 4 cores, or will it be the A9x clocked at a higher speed and with more RAM.
My CPU and GPU performance assumptions are as follows: A8 < A8X < A9X and A8 < A9 < A9X (the existence of the A9 and A9X are implied).

I've been thinking about this for a while. There have been conflicting rumors about whether or not the iPad Air will be updated this year, some rumors say yes, others say no (see near the bottom of this page). That's not the case for the iPad mini—from what I remember, every rumor about it this year has claimed that it will actually get an update this year. This situation reminds me of the iPad mini rumors last year, so I have a suspicion that the iPad Air 3 could be at most a minor update, similar to the iPad mini 2 to iPad mini 3 update.

If so, then I would expect the iPad Air 3 to retain the A8X, which I think would imply one of these possibilities:
  1. It is not true that A8X < A9 in both CPU and GPU performance, and the A9X is not suitable for the iPad Air.
  2. The A8X was the preferred SoC for the iPad Air 3 for reasons other than performance or power.
If the A9 falls short of the A8X in CPU or GPU performance, then it probably won't be a good A8X successor. An A9 with three CPU cores or two CPU cores with large per-core improvements and a GT7600, which I think isn't particularly far-fetched, should be able to meet or exceed the performance of the A8X. But I don't think that's the whole picture. According to what I've read (from posts on Beyond3D forums as well as this), the 14/16 nm nodes are supposed to be very expensive, and that might affect the choice of SoC in the iPad Air 3. For example, if the A9 costs more than the A8X and doesn't have a significant performance increase over it (say, ≤ 20% in both CPU and GPU), then I am not sure if it is worth it for Apple to update the iPad Air with an A9, especially if the iPad "Pro" rumors are true.

Question: what do you folks expect the cost of an A9 with 2 CPU cores and 4 Series 7XT GPU clusters to be (assuming the CPU cores don't grow by a lot)? What about an A9 with 3 CPU cores and 6 Series 7XT GPU clusters?

I suspect that the A9X will be very powerful and be designed with the iPad "Pro" in mind. My current guess is 4 CPU cores and a 12-cluster Series 7XT GPU (maybe 16 but that takes a bit more optimism than what I have). I also expect one of Apple's areas of focus for the iPad "Pro" to be on new apps that are not feasible on an iPad Air 2 but are on a 13" tablet with 1.5-2x the computing power.

According to the rumors so far, the iPad "Pro" will have these features that the iPad Air 2 does not have.
  • 1.8x the display area and point resolution (1366x1024 vs. 1024x768)
  • A more accurate touchscreen
  • Optional stylus
  • Stereo speakers on the top and bottom
  • Force Touch
  • USB-C port [maybe]
  • A larger, wider onscreen keyboard that resembles a desktop keyboard
The following features are not necessarily rumored, but I think they are sensible predictions.
  • A9X SoC with a decent performance jump over the A8X
  • It will be long enough for a full-size physical keyboard (the display seems a tiny bit too small for a full-size onscreen keyboard but it might be passable)
Many of these features can in principle be brought to a iPad Air-sized device as well. The important observation is which features can't:
  • A9X SoC with a decent performance jump over the A8X
  • 1.8x the display area and point resolution
  • A larger, wider onscreen keyboard that resembles a desktop keyboard
  • It will be long enough for a full-size physical keyboard
I expect the iPad Air of 2016 or 2017 to match or exceed the performance of the iPad "Pro," so the performance advantage of the iPad "Pro" over the iPad Air of the same year is only an advantage in the long-term for the tasks whose minimum requirements are also increasing over time at similar rates.

Perhaps Apple will leverage the larger display area and higher resolution to open the door to more "professional" software and other apps that show lots of information on a display. 1366x1024 is in laptop territory by the numbers, and while a straight comparison is not possible due to factors such as touch target sizes, some pieces of information are not particularly dependent on the input method of the device it's on. I would also assume that a fair number of these apps could benefit from a full-size keyboard or USB-C, which justifies the latter on the iPad "Pro" (the former follows from the iPad's size). Now how does this relate to the SoC? Well, if Apple goes in this direction, then I think they will spec the A9X so that it is powerful enough to handle any software, new or existing, that the iPad "Pro" will be running, so all the way from note taking to advanced video editing and high-resolution gaming.

Those are my thoughts. But I was way off on my A8 predictions so…. :p
 
As the expected launch gets closer, there have been quite a lot of leaks of supposed iphone 6s parts, and some ipad mini parts. However I don't think I've seen a single leak of a full size ipad part, or indeed ipad pro. This leaks me to be sceptical as to a launch of anything other than a new mini at this sept event. I hope I'm wrong, as I would like a bigger ipad.

I suspect that given the relative down turn in volumes of the ipad, that one soc will cover both ipad an ipad pro. It probably makes more sense for apple to design separate highly targetted socs when you have big volumes. So possibly, they'll use one soc, with some CPU/GPU cores switched off on the regular ipad.

According to the rumors so far, the iPad "Pro" will have these features that the iPad Air 2 does not have.
  • 1.8x the display area and point resolution (1366x1024 vs. 1024x768)

The regular ipad has been at 2048x1536 for ages.
 
Last edited:
1024x768 is still the virtual/point resolution though. Apple have decoupled the hardware resolution from the software one for HiDPI devices since their introduction (and for all devices really, just they might be 1:1 mapped). There's a handy infographic somewhere, I'll try and dig it out.
 
Back
Top