Apple A8 and A8X

I was wondering if the Samsung 14 nm process having a lower number than the TSMC 16 nm FF+ process meant that it was better in some way.

Well, it's obviously better by 2nm, but worse by one +.



OK, in all seriousness, process names are merely marketing. You can't really tell how these processes compare with each other without knowing a whole lot of characteristics, such as contacted gate pitch, SRAM cell size, as well as things like clock/voltage-scaling, capacitance, defect rate, etc., which tend to all depend on one another, plus other things. You may find this old article useful: http://www.realworldtech.com/iedm-2010/

Realistically, unless you have a chip to build in large quantities and you contact both foundries to get detailed information, you're not very likely to get a real answer to "which process is better?". The amount of business each foundry gets may offer some clues, though.
 
Last edited:
Samsung does have a density and power advantage.

That is what I had read as well..thanks for confirming it. This is 14LPE (Low Power Early) we are talking about right? They have an upgraded version called 14 LPP (Low Power Plus) available later in the year as well.

I read an article just yesterday on how Samsung was able to catch up to TSMC so fast. I dont know how much of it is true but it was certainly an interesting read. A bit off topic but here is the link in case anyone is interested - http://english.cw.com.tw/article.do?action=show&id=14895
Don't let anyone from Intel catch you saying that :runaway:

:LOL:
 
That is what I had read as well..thanks for confirming it. This is 14LPE (Low Power Early) we are talking about right? They have an upgraded version called 14 LPP (Low Power Plus) available later in the year as well.

I read an article just yesterday on how Samsung was able to catch up to TSMC so fast. I dont know how much of it is true but it was certainly an interesting read. A bit off topic but here is the link in case anyone is interested - http://english.cw.com.tw/article.do?action=show&id=14895

Woah !
 
I'm posting here because I don't want to derail the Samsung thread.

I hope Apple make an iPad Pro, where they allow the SoC die area to grow substantially. It would be very interesting to see what could be done with a mobile architectural base, FinFETs, and 200mm2 or so die area.
Not likely to happen, of course.
Suppose that an 12"-13" iPad "Pro" was released last year along with the iPad Air 2. Would it have made sense to target the A8X at the iPad "Pro," with 4 CPU cores and 12 GPU clusters (pretend that Series 6 can scale past 8 clusters, or take option 2: 8 clusters at ~600 MHz), and cut it down to 3 CPU cores and 8 GPU clusters (option 2: keep the 8 clusters and reduce the clock) for the iPad Air 2?
 
S7XT cores scale up to 16 clusters if needed and are supposed to increase clock for clock and cluster for cluster 60% in performance compared to Series6XT. If true there's no absolute necessity to scale beyond 8 clusters.
 
I can't really see the retrospective scenario iMacmatician. However, I'm quite curious about what a hypothetical iPad Pro SoC might look like this autumn. There have been references to x3 resolutions, which would mean 3072x2304 pixels, or 2.25 times as many as the iPad Air2.

I assume that Apple will utilise either Samsung 14nm LPE or LPP, or TSMC FF+ both of which will allow some moderate density improvements vs. TSMC 20nm. A very straightforward scenario would be for Apple to stay triple Cyclone with tweaks, and exchange the GPU for Series7XT GT7800 with 8 clusters as you suggest Ailuros. It will perform nicely, particularly if coupled with a 128-bit interface to DDR4-3200. It would probably allow a bit higher clocks all around, at similar power draw as the A8x.

However, that would pretty much be what I would expect for an iPad Air3. If I were Apple, and introduced a new class of iPads, I might want to aim higher. Larger battery and body/heatsink might allow somewhat increased power draw, and being a tier up in the hierarchy would allow higher SoC costs. Furthermore, they might want to truly emphasise that tablets are not only for web surfing but are much more capable than that. Something aiming to challenge the surface 3 Pro, which would be the inevitable reference point if Apple chooses to use the "Pro" moniker. (plus, the A8x already crushes CherryTrail 14nm Atoms). If so, they would probably want something like 25% or more improvement in single thread CPU performance, four cores, and might even go for the GT7900 (16 cluster) GPU, which, again, if married to a 128-bit DDR4 memory interface should perform very nicely indeed. It would challenge Intels best efforts with Broadwell in terms of CPU, and exceed it in graphics. The CPU single thread performance improvement could probably in large part be achieved with clocks. Having faster main memory, perhaps shaving cycles off cache latencies, increasing the size of the L3, tweaking branch prediction, et cetera would help single thread performance to desired levels. Having four real cores would help multithreaded performance comfortably outperform Intels dual core CPUs.

This high-end scenario seems to be doable on the the new process nodes, but the die area would probably be somewhat larger than the current A8x.

Of course, a counter-argument for such a chip is that it would arguably be too expensive for the regular 10" iPad, and that the 12"+ volumes can't be trusted to merit its own SoC. Which unfortunately seems reasonable. If on the other hand the 10" and 12"+ share the same SoC, I still have a feeling it could turn out to be a bit more ambitious than A8x with an updated 8 cluster GPU.

[MOD: Your friendly editor has just added some breaks between paragraphs :) ]
 
Last edited by a moderator:
I'm not sure but it's my understanding that the GT7900 is DX11 only; if yes and there isn't a DX10.0 variant I severely doubt Apple would want to bite the >50% additional transistor count for something they won't have any use for, for quite some time to come. I also wouldn't expect anything as large as a 7900 before 10nm or an equivalent process. Considering alone that it has 32 TMUs it sounds too far fetched for anything 16/14FF for a hypothetical 12-13" device that shouldn't also weigh as much as concrete cement brick.

One theoretical solution if Apple truly wants the large Pad to differentiate (if it even exists after all as a project....) would be to take a GT7600 and just mirror it for 12 clusters. My gut feeling however so far tells me that if a GT7800 clocked at say 600MHz will deliver =/> performance than the X1 GPU, that they will NOT aim for anything higher than that. IMHO a potential =/>12" user will be more happy with a slightly smaller GPU but N% more battery lifetime instead AND of course X sustained performance :p
 
Well. I'm inclined to agree, mostly because I don't think Apple would give an iPad Pro a SoC of its own. The A5x had a substantially larger die than the A8x though, so it is not completely inconcievable that they would let SoC area increase if they want to push a "professional" iPad.

As far as feature set of the GPU is concerned, Apple controls Metal, so anything that's there could be accessed by developers if Apple so desires. You call supportsFeatureSet: of your MTLDevice which returns an MTLFeatureSet. The mechanisms are already in place to make use of new GPU features a reasonably smooth process.
 
The DX10.0/GPU in A8X is under 20nm at 38+mm2. That die area doesn't help because they just mirrored a quad GX6450 but let's ignore that for a second. I'd estimate that it would blow up in die area nearly to 60mm2 under 20nm in theory for an 8 cluster DX11.2 7XT part. Since they're scaling just clusters die area shouldn't scale linearly upwards, but are you sure the bullet for 16 DX11.2 clusters is already "biteable" under 16/14FF? If yes it would mean a ULP SoC GPU roughly as big as the entire Exynos 7420 on estimate :p
 
As far as I know Apple is working on a in-house GPU. I don't know if it will be in the A9 or beyond, but look at their job listing specifically in Austin and Orlando. Over a year ago, the company I work for lost two RTL designers to Apple to their GPU team so they've been working on it for a while now.
 
iPad sales are flattening.

Will they invest heavily if sales aren't growing?
My personal feel is that Ipad is finally at a good place with it's OS and ecosystem. Previously it was fairly limited if not leveraged as a reader/watcher, but it can be so much more now. With the right setup it can be an effective productivity tool, though, a heavy amount of scripting knowledge is required; for most casual users it still isn't quite there yet. That being said, Apple needs to decide if they want to continue down this route of making it more functional, or to stop now, and have the 'new' Macbook take that place.

I don't necessarily know what Apple is going to do, they've held onto to this separated IOS/OSX ecosystem for some time, and it looks like they intend to for a bit longer.
 
iPad sales are flattening.

Will they invest heavily if sales aren't growing?

No tablet sales are flattening in general, which in essence is just an idiotic analyst forecast were phablets are counted separately. Else anything above 7" doesn't see any worthwhile increase. If Apple can sustain for the next few years their tablet sales volumes it doesn't necessarily mean that they'll sell less than before or that past investments were in vain in that regard. As Entropy already said one past tablet SoC was over 160mm2 big for which timeframe their tablet sales had probably been smaller than they are today.

Long story short with Apple's typical profit margins there's enough headroom for investments. If Apple shouldn't have any then any other tablet maker should pack up and go home also.
 
Back
Top