Apple A9X the new mobile SoC king

[Eug]

Did anyone doubt IDC?

Nope. Here's what he said:

http://forums.anandtech.com/showpost.php?p=37716711&postcount=34

Considering that the majority of the A9s tested thus far are TSMC, I'd say that this was pretty spot on.

Earlier in this thread, I posted an article from a tech publication that even back then in 2013 said that TSMC was getting 60-70% of the volume for A9, whereas 30-40% of the volume would go to Samsung.

IDC also claimed that because Samsung yields were low, Apple couldn't release the phone earlier. That makes little sense, because Apple released the phone when they always release the phone, in September/October. They've done that since the 4S.

And this is why Apple had to wait to release the iphone 6S versus releasing it months ago (when Samsung 14nm HVM was obviously available), as TSMC had practically no substantial 16FF+ capacity online until about 2 months ago. So the tradeoff was to wait for TSMC's high-yielding 16ff+ capacity to come online, or take Samsung's much lower yielding 14nm capacity and launch earlier in the year.

So, it sounds very much as if the conclusions in his posts were big extrapolations from a little bit of information.

 

[ShintaiDK]

GloFo failed flat out. And Samsung couldn't deliver what they should.

 

[Fjodor2001]

GloFo failed flat out.

Were they ever in the race for early 14 nm production? I thought it was generally known that they'd be following in the footsteps of Samsung, since they're copying their process tech.

Also, I'm seeing A10X on Samsung 10 nm in 2016? Could that really be true? Intel is gunning for 10 nm in late 2017 or 2018 IIRC (And yes I know their 10nm might be a bit better, but still... so down spawn off into an OT discussion on that. )

 

[Arachnotronic]

Were they ever in the race for early 14 nm production? I thought it was generally known that they'd be following the in footsteps of Samsung, since they're copying their process tech.

Also, I'm seeing A10X on 10 nm in 2016? Could that really be true? Intel is gunning for late 2017 or 2018 IIRC (And yes I know their 10nm might be a bit better, but still... so down spawn off into an OT discussion on that. )

Unlikely. The foundries have publicly said that they plan to start volume production on 10nm in late 2016. There is zero chance that we see 10nm A-series processors in 2016, IMO.

 

[pepone1234]

TSMC claims their newest version of FinFet produces a transistor that is 10% faster than Samsungs latest FinFet.

I don't think we are looking at TSMC 16nm FF+ because this is mostly their first mass produced thing with their new process. I think this could be 16nm FF without that plus.

 

[Phynaz]

Looks like IDC was right.

Yep. And some posters in this thread still owe him an apology.

 

[Space69]

What are you on? There's no "constant rewriting." You change a few things around to fool the benchmark cheating, assuming any exists.

If you experience any changes in the behaviour you would need to decompile and analyze the new code - you do not have symbols and debug information. Afterwards you create a solution - you'll have to compile and test on all other supported platforms - Android, OSX, Win and Linux - before release.

 

[Space69]

The point is, these optimizations are accepted because a benchmark is supposed to closely simulate real-world performance. Furthermore, all developers have access to them equally.

See above.

Then you're not able to benchmark different architectures against eachother - I can understand that you accept the use of GPGPU resources on one platform and CPU on another for benchmarking different architectures. I don't.

 

[Nothingness]

Then you're not able to benchmark different architectures against eachother - I can understand that you accept the use of GPGPU resources on one platform and CPU on another for benchmarking different architectures. I don't.

Why? It's a part of the SoC. Or should we be sure AVX isn't used too? After all you don't find it on all CPU from Intel.

I see what you mean and agree in a way, but if that GPU targetting improves end user apps, then its use is acceptable.

 

[Nothingness]

If you experience any changes in the behaviour you would need to decompile and analyze the new code - you do not have symbols and debug information. Afterwards you create a solution - you'll have to compile and test on all other supported platforms - Android, OSX, Win and Linux - before release.

My understanding is that in this case you wouldn't have access to the compiled binary and so won't be able to do any reverse engineering (or perhaps on jailbroken devices?).

 

[coercitiv]

IDC also claimed that because Samsung yields were low, Apple couldn't release the phone earlier. That makes little sense, because Apple released the phone when they always release the phone, in September/October. They've done that since the 4S.

1st gen: June 29, 2007
3G: July 11, 2008
3GS: June 19, 2009
4: June 24, 2010

Always... such a restrictive word... sometimes ()

 

[Eug]

^^ Hence my post saying since the 4S.

Ie. The last 5 releases. Also, I see you made no comment on the article from 2013 specifically saying that in 2015 Apple would give 60% of its A chip business to TSMC.

 

[coercitiv]

^^ Hence my post saying since the 4S.

Just because you mention that in your post doesn't make it a better crafted statement. Your post made it sound like they "always" launched in September/October. Dates paint a different picture: 50% of past launches were made earlier than September, and the 4S launch was marked with lots of rumors and speculation regarding the reason for the "delay".

"Little reason" looks a bit bigger now, don't you think?

 

[catinthefurnace]

1st gen: June 29, 2007
3G: July 11, 2008
3GS: June 19, 2009
4: June 24, 2010

Always... such a restrictive word... sometimes ()

Seriously!? Did you not read the last sentence in his post that you quoted!?

 

[Eug]

Just because you mention that in your post doesn't make it a better crafted statement. Your post made it sound like they "always" launched in September/October. Dates paint a different picture: 50% of past launches were made earlier than September, and the 4S launch was marked with lots of rumors and speculation regarding the reason for the "delay".

"Little reason" looks a bit bigger now, don't you think?

No, not at all, and your internet handwaving to try to make it sound suspect isn't working. Sure, perhaps I could have crafted it better, but I very specifically said it has been that way since the 4S, which is the last 5 releases.

Furthermore, I count yet another post of yours where you specifically ignore the 2013 industry article, which indicates this was the plan all along. Almost 2 years before the iPhone 6s/6s+ comes out, the article states that TSMC won a contract to get 60-70% of the new 2015 iPhone business, while Samsung won a contract to get 30-40% of the new 2015 iPhone business. And you know what? That's just what happened, according to the 2500+ sample of iPhones that have been tested thus far.

This makes it abundantly clear that Samsung didn't suddenly lose a whole bunch of business in the last few months because their 2015 yields were low. They never had that business in the first place.

It really just sounds like someone got the numbers of TSMC and Samsung from somewhere, and then a whole bunch of speculation was thrown out on top of that. I won't say where that speculation came from, but it should be pretty clear that is just that, and wrong.

 

[catinthefurnace]

Just because you mention that in your post doesn't make it a better crafted statement. Your post made it sound like they "always" launched in September/October. Dates paint a different picture: 50% of past launches were made earlier than September, and the 4S launch was marked with lots of rumors and speculation regarding the reason for the "delay".

"Little reason" looks a bit bigger now, don't you think?

"Made it sound like"????

He PLAINLY SAID since the the 4S.

 

[lopri]

This site has close-up shots of TSMC-made A9. They require registration, though, and I do not want to infringe any copyrights..

http://www2.techinsights.com/l/8892/2015-09-28/zxx9c

The report is a detailed structural analysis (LDSA) of the Apple A9 process fabricated using TSMC’s 16 nm finFET process.

The A9 is fabricated using an 12 metal (11-Cu, 1-Al), 16 nm finFET CMOS process. The device features approximately 33 nm long metal gate finFET transistors having a 90 nm minimum contacted gate pitch; the same as used by their 20 nm HKMG process released in the summer of 2015.

The reported results are derived from scanning electron microscopy (SEM), transmission electron microscopy (TEM), energy dispersive X-ray spectroscopy (TEM-EDS), electron energy loss spectroscopy (EELS) and spreading resistance profiling (SRP).

 

[dark zero]

I want a benchmark comparison between A9 TSMC and A9 Samsung... I wonder how much the Samsung A9 is stronger than TSMC one...
Because there might be difference in performance between 14 nm and 16 nm.

 

[catinthefurnace]

I want a benchmark comparison between A9 TSMC and A9 Samsung... I wonder how much the Samsung A9 is stronger than TSMC one...
Because there might be difference in performance between 14 nm and 16 nm.

And it might be the opposite of the result you expect.

 

[Thanatosis]

Rumors were that TSMCs process was "highly impressive" to Apple, and has superior power characteristics to Samsung's 14nm node. It is also pretty much established that Samsung's 14nm node is the product of industrial espionage in a way that no other node has ever been before. A TSMC engineer left Taiwan two years ago for Korea and brought with him all the work up until that point.


Samsung is currently being sued for this, but even when they lose the suit years later the fines will never come close to the amount of $ that stolen tech will have earned them.

 

[catinthefurnace]

Samsung is currently being sued for this, but even when they lose the suit years later the fines will never come close to the amount of $ that stolen tech will have earned them.

That seems to essentially be the cornerstone of their corporate philosophy.

 

[coercitiv]

Furthermore, I count yet another post of yours where you specifically ignore the 2013 industry article, which indicates this was the plan all along. Almost 2 years before the iPhone 6s/6s+ comes out, the article states that TSMC won a contract to get 60-70% of the new 2015 iPhone business, while Samsung won a contract to get 30-40% of the new 2015 iPhone business. And you know what? That's just what happened, according to the 2500+ sample of iPhones that have been tested thus far.

We were talking about the possibility of Apple wanting to launch their new iPhone in the summer months. You didn't reject that possibility based on the foundry split, but rather on the sole argument that Apple releases phones when they always release phones, and that is in September/October. Ever since the 4S. But ever until the 4S Apple did launch their phones in the summer and they did that for a reason, which in turn means they may very well consider gradually moving back to that schedule when certain conditions are met.

I can definitely see your point of view: Apple is purposely launching in September every year, but you should still keep in mind something happened on the 4S release that might have set them on this path, and it's possible (but admittedly not highly probable) they might change that in the future.

Anyway, we both made our points, worst case we can agree to disagree.

Are you freaking serious!? Did you not read the last sentence in his post that you quoted!?

Wow, I find it hard to believe that you believe what you're saying.

Good god almighty what is wrong with the world?

Welcome to the forums. Please be civil.

 

[raghu78]

Looks like IDC was right.

Sorry but too many people are making elementary mistakes. 2500 samples from 13 million units sold in 3 days. Thats a 0.0192% sampling rate. There is no way you can arrive at a correct conclusion on TSMC and Samsung ratio with such a low sampling rate. btw idc claimed its all TSMC due to woeful yields at Samung. So he is still wrong.

http://blogs.barrons.com/asiastocks...save-tsmc-why-goldman-still-likes-this-stock/

"TSMC now expects FY15 sales to be up slightly less than 10%, lower than previous guidance of well over 10%. 2H sales are now expected to contract 3-5% HoH compared to previous guidance of better 2H than 1H. These imply a third consecutive misguidance in FY15.

We expect the current inventory digestion to be largely over in 4Q15 but a recovery in 1Q16/FY16 may not necessary materialize as demand remains uncertain. Besides, TSMC is losing QCOM’s S820 orders (high-end AP to Samsung), which typically ramp in 4Q/1Q to support new Android premium models. Also, TSMC remains the second source for Apple’s A9 which is subjected to higher risk should demand falter. Maintain SELL."

btw TSMC being second source indicates they are not the one supplying higher volumes.

 

[Arachnotronic]

Sorry but too many people are making elementary mistakes. 2500 samples from 13 million units sold in 3 days. Thats a 0.0192% sampling rate. There is no way you can arrive at a correct conclusion on TSMC and Samsung ratio with such a low sampling rate. btw idc claimed its all TSMC due to woeful yields at Samung. So he is still wrong.

http://blogs.barrons.com/asiastocks...save-tsmc-why-goldman-still-likes-this-stock/

"TSMC now expects FY15 sales to be up slightly less than 10%, lower than previous guidance of well over 10%. 2H sales are now expected to contract 3-5% HoH compared to previous guidance of better 2H than 1H. These imply a third consecutive misguidance in FY15.

We expect the current inventory digestion to be largely over in 4Q15 but a recovery in 1Q16/FY16 may not necessary materialize as demand remains uncertain. Besides, TSMC is losing QCOM’s S820 orders (high-end AP to Samsung), which typically ramp in 4Q/1Q to support new Android premium models. Also, TSMC remains the second source for Apple’s A9 which is subjected to higher risk should demand falter. Maintain SELL."

btw TSMC being second source indicates they are not the one supplying higher volumes.

What makes you think the Goldman Sachs analysts are correct?

p.s. seems like you were wrong in claiming that Samsung -> 6s, TSMC -> 6s+

 

[EightySix Four]

Sample is up over 12,000 units now: http://demo.hiraku.tw/CPUIdentifier/