Not All iPhone 6s Processors Are Created Equal (itworld.com)

itwbennett writes: Apple is splitting the manufacture of the A9 processor for its iPhone 6s between TSMC (~60%) and rival Samsung (~40%) — "and they are not created equal," writes Andy Patrizio. For starters, Chipworks noted that Samsung uses 14nm while TSMC uses 16nm. A Reddit user posted tests of a pair of 6s Plus phones and found the TSMC chip had eight hours of battery life vs. six hours for the Samsung. Meanwhile, benchmark tests from the folks at MyDriver (if Mr. Patrizio's efforts with Google Translate got it right) also found that the Samsung chip is a bigger drain on the phone's battery, while the TSMC chip is slightly faster and runs a bit cooler. So how do you know which chip you got? There's an app for that.

Comments

Too little, too late

[akahige]

More to the point, how can you find out which chip the phone has before buying it?

Re:Too little, too late

[Grishnakh]

Simple: don't buy it at all. If a company is going to play shenanigans like this where products marketed with the exact same name and part number are significantly different and it's just a luck-of-the-draw ass to whether I get the good one or the crappy one, I'm just not going to buy their product at all.

Re:Too little, too late

[soft_guy]

The fact that they ship an improved version every year or so is NOT the issue here. Seriously if the new version is not a big enough improvement over the one you have, you don't need to buy it. You can keep your phone for 2 years, or 3 years, or however long you want. That is hardly "pushing it through the throats of customers". You have to have a major victim mentality to think that. I do agree though that shipping non-equivalent versions of the processor is a big deal. That's not okay.

Re:Too little, too late

[gstoddart]

I'm sorry, but who is pushing a new iPhone "through the throats of customers"??

You are completely free to not fucking buy one.

Did you know that car makers push out a new version, only slightly different, annually? Companies who make golf clubs, also push out new versions at least annually. And companies who make TVs, they also do this.

If customers buy a new expensive phone every year or two, don't blame the vendor. Free will doesn't stop just because you've bought a product.

Re:Too little, too late

[DarkOx]

I don't understand why people freak out when a tech vendor releases a new model, as if they are forced to buy it or the one they have is suddenly going to explode. I do think some large vendors are guilty of abandoning support for their legacy products a bit to quickly. Nobody gets all nuts about the fact the Chrysler/Ford/GM/Honda/VW/Mercedes/etc bring out new models every year; often with slight improvements, usually with other changes you may or might not like.

Re:Too little, too late

[tripleevenfall]

A vendor making a mid-model year substitution for a better product seems like a benefit to me, not a detriment. Instead of nobody getting the improved version, at least some people do.

Re:Too little, too late

[Anubis IV]

As more data is coming to light, it's sounding like the differences are smaller than first estimated and were likely exaggerated as a result of synthetic benchmarks not accurately modeling real-world performance. Specifically, while it does sound like TSMC's chips are performing far better than Samsung's in the synthetic benchmarks (e.g. Geekbench's battery tests), MacRumors has some followup on the topic, indicating that in the real world tests that are ongoing, the results so far appear to be much closer between the two models.

It sounds like the synthetic benchmarks may be slamming a part of the processor that TSMC has optimized better than Samsung, but that in real-world performance, that part is used far less frequently than in the synthetic benchmark, meaning that the results from the synthetic benchmark may not accurately model real-world performance.

Re:Too little, too late

[UnknowingFool]

Did Apple Spec out the correct specs to Samsung and they made a cheap knockoff, after sending a batch that seems to meet initial QA, in a very German style. Or did Apple know about/agree to giving different quality products.

There are both done on slightly different processes and it seems there is a difference that should not be there. It may be that somewhere in the Samsung process (masking, lithography, etc), a difference is significant enough to cause this battery issue. Or that something about Apple's design (or chip design in general) is affected by the step down to 14nm that isn't noticeable in 16nm. Remember that chip features are starting approaching the limits where problems occur that require design changes like multi-gate which did not occur at larger feature sizes.

Re:Too little, too late

[UnknowingFool]

If sourcing substantially different parts from different vendors is necessary to meet production volume, then they need to have different part names and model names for these products.

Let's start with the first thing: These parts are not supposed to be substantially different. They are the same design but at a smaller feature size. The fact that they are is a problem and Apple will have to get with Samsung and TSMC to figure what is the issue. Second, different sources parts are known internally but not externally. After all, does Intel rename a Broadwell Core i7 differently when it comes from Oregon or Arizona or Ireland? No. There is a part number that tells where the chip was made and you as a customer don't know where it came from when you order it from Newegg or Micron or wherever.

This isn't a case of having resistors or capacitors from different manufacturers, something that won't affect performance in any measurable way, this is a case of having two completely different CPUs, with very different performance from the two.

How is an dual core A9 from Apple a "completely different CPU" than an dual core A9 from Apple. They are the exact same design by Apple. If you feel that makes them "completely different", did you lecture Microsoft when they switched Xbox processors? From what I remember IBM Xenon processor was shrink reduced from 90nm to 65nm to 45nm. These are all "different" CPUs to you?

6h vs. 8h in a power-consumption test is a huge, huge difference.

And if it's true, Apple will have to look into why.

Intel sells CPUs all the time which are very similar, but have performance that differs to that extent: they use completely different part numbers to describe these parts.

The problem with this comparison is that a Core i7 is not the same as a Core i5 with actual differences like L3 cache size, TDP, clock speed, etc. and these come from different designs.

Re:Too little, too late

[sg_oneill]

Apple has been generally pretty good with that. Older iPhones will still run newer software, although in some cases its debatable if its actually a good idea to do so, if the software is written under the assumption of a more performant processor. At least with the laptops, my Macbook 2011 is running the latest and greatest OSX at a cracking pace, and my GFs iphone 5 is fully updated and running well.

Re:Too little, too late

[AmiMoJo]

They have a history of doing this too, because like like to have two suppliers competing. Most of the time no-one cares, but for example a few years back they had all those dodgy "retina" LCDs on their laptops. There were two suppliers, LG and Sharp, and all the LG ones were prone to ghosting while the Sharp ones were fine.

The only thing customers could do is keep taking the laptops back when ghosting appeared and hoping that the replacement had a Sharp LCD. I think that's what upset people, the way it was handled. On the other hand I can appreciate that Apple probably didn't want to replace millions of LCD panels if they could avoid it.

This is pretty shitty though, because the lesser of the two CPUs isn't defective and thus the customer can't swap the phone for a better one. They just have to live with reduced battery life and performance on their very expensive new shiny. It will be interesting to see what Apple do. For example, if they ban apps that detect which CPU you have then people will endlessly speculate about why their battery life is crap and keep worrying about it, but if they don't people will demand exchanges. Could affect resale price too.

Re:Too little, too late

[AmiMoJo]

Yep, it's most likely a problem with the design. Shrinking from 16nm to 14nm isn't simply a case of scaling your design files by 87.5%, you have to make separate ones. You can carry over most of the high level design and layout, but the computer has to re-synthesise the detailed transistor structure, you might have to use different cache memory, different power and voltage management devices etc. 14nm is a different process, it's not just a slightly better focus on a lens or something.

So the two were never going to be exactly the same, and chances are it's just that the 16nm design is a bit better optimized. Could be that it makes better use of the materials used, could be that the computer did a better job on synthesis, could be a number of things. I really doubt that Samsung sabotaged it though.

Re:Too little, too late

[DerekLyons]

But who is playing shenanigans Samsung or Apple.
Did Apple Spec out the correct specs to Samsung and they made a cheap knockoff, after sending a batch that seems to meet initial QA, in a very German style. Or did Apple know about/agree to giving different quality products.

There's a third possibility that should not be discounted out of hand - Samsung meets the specification, while TSMC exceeds it. Without access to internal information, it's hard to tell what's going on behind the curtain and all too easy to leap on the 'obvious' conspiracy.

Of course, the various mega corps routinely indulge in behavior that makes conspiracy theories not all that far fetched...

Re:Too little, too late

[Anubis IV]

MacRumors didn't do any testing at all. They're just compiling lists of tests that others are doing, in an effort to get a sense for whether or not there actually are large differences. And the A9 chips being provided by Samsung and TSMC are SoCs, not just CPUs, so it makes perfect sense that they'd be running benchmarks that include video processing and other non-CPU-bound tasks, given that the A9 is responsible for those as well.

Moreover, given the variance in performance that can occur within chips from even a single manufacturer, it's no surprise that there will be variation between the two models. If there weren't, it would be a news story. As such, the important questions to ask are:
1) Is the difference between the two broadly reproducible (i.e. is Samsung consistently behind), or is it this just an anecdotal case involving a single low-performing Samsung chip being compared to a single high-performing TSMC chip?

2) Given the variation, do either of them fall below the specs provided by Apple?

We don't have enough data yet to answer #1, but, again, as more data is coming to light, it's sounding like things are not so lopsided as the initial reports indicated. TSMC may have a slight edge, but it's not anywhere in the ballpark of what was being reported earlier. As for #2, by all indications, the answer is "no, neither of them fall below Apple specs". Which is to say, some people may win out on the luck of the draw and get a phone with a chip that performs better...which was already the case anyway, since chips are never perfectly identical in their performance. All a manufacturer will do is guarantee that the performance falls within a certain range, so some will always perform better than others, even when built using the same process from the same manufacturer.

Re:Too little, too late

[Aaden42]

Apple doesn’t “send down” any updates. You’re free to take or leave any OS update you like. They’re remind you a bit, but nobody is forced to upgrade the OS they have on their phone right now.

And do you *seriously* think Apple releases updates to “actively try to fuck up older but functioning hardware”? Paranoid much? Yes, some updates have made older hardware work less well. Other updates have improved long standing issues on older hardware. That’s the nature of software development. It’s not a good thing, but it’s a far distance between “didn’t test it as much on three-year-old hardware” and “let’s intentionally add this bug to make the old phone flake out.”

Only seen in specific benchmarks

[hattig]

http://www.macrumors.com/2015/...

As suspected from early results yesterday, the takeaway from Morrison and Evans' videos today seems to be that while intense cases like synthetic Geekbench tests designed to push devices to their limits can reveal significant differences in battery life between devices using the two chips, real-world impacts are much smaller and are likely to be unnoticeable to many users.

Re:Battery Life

[TWX]

I've found that battery life on standby is very much dependent on carrier accessibility. My employer's campus has a power distribution station on the East side and is ringed on the North, South, and West sides by power lines that reach the station. We get very poor signal strength and my old Galaxy SII is lucky to survive the eight hour shift on battery if I'm at the office all day, even on standby.

Contrast to at home, where that city mandated all infrastructure be buried, and the power lines are only for neighborhood final distribution as opposed to regional distribution, and my phone can go a whole weekend on standby.

One no sim??

[Anonymous Coward]

How come

On the chinese test, the Samsung has an extra app installed on it (see the screen of the doc).
And on the Reddit users test, the TSMC has a sim card installed, the Samsung not.

Really would it have killed them to keep the same spec for each?

16 nm vs 14 nm

[chris200x9]

It's kind of interesting the CPU built on a larger process is faster, cooler, and has less power draw.

Re:16 nm vs 14 nm

[TheRaven64]

I'm not particularly familiar with either company's process, but it's been a couple of generations since you could actually make meaningful comparisons based on the quoted nm size, because everyone has different smallest features that they measure when deciding that they are Xnm. That said, we passed the end of Dennard scaling a long time ago. You'd expect the same chip to be consuming about as much power, be slightly more able to dissipate the heat. It may also have less leakage, though that depends on a number of other factors.

Re:Battery Life

[Anonymous Coward]

Power lines cause very little interference to cell phones. It is very likely your office has eco-friendly windows. They have a very thin metal coating which greatly attenuates the signal. It is the same at my company. A part of our bulding recently had new windows (and some other eco-tweaks) installed and the signal there is almost gone, returning to almost full strength when the window is opened.

Samples sizes of 1

[radarskiy]

"tests of a pair of 6s Plus phones"

You can't argue with the statistical validity of that analysis... because there isn't any.