TSMC catching up with Intel on leading edge process nodes

[jpiniero]

Samsung's probably playing more games with the node names. I wish they'd stop.

 

[Fjodor2001]

Samsung's probably playing more games with the node names. I wish they'd stop.

So no answer to any of my questions? And you're still sticking to your claims despite that, with no sources to back anything of it up?

 

[superstition]

the market (gamers mostly) continued to pay for new hardware at much higher prices.

This is why the industry is salivating over 8K. As long as pixel count continues to get an undue amount of attention it's easy enough to induce people to spend a lot on GPUs.

The fact is that a game at 1080 res game could look like reality and be more interesting than a 4K game where everything looks video game quality.

And there are other factors that would make a bigger difference once you're past 1440, like contrast ratio and color gamut — both of which greatly lag behind the capabilities of human vision. Contrast, at least, is being addressed by OLED but it remains to be seen if it will be safe enough for burn-in on the desktop, especially with the obsession over tiny pixels.

 

[zentan]

Samsung's probably playing more games with the node names. I wish they'd stop.

Well,Idontcare and many others did say that it would happen.We did see this with Intel 14nm vs TSMC 16nm vs samsung 14nm.It seems it would be the case with newer nodes.Probably samsung's 10nm would be somewhat better than Intel's 14nm but might be lagging behind by a considerable amount when compared with Intel's 10nm. But we will have to wait and watch for that.Things are a bit far away to be crystal clear.

 

[Arachnotronic]

Well,Idontcare and many others did say that it would happen.We did see this with Intel 14nm vs TSMC 16nm vs samsung 14nm.It seems it would be the case with newer nodes.Probably samsung's 10nm would be somewhat better than Intel's 14nm but might be lagging behind by a considerable amount when compared with Intel's 10nm. But we will have to wait and watch for that.Things are a bit far away to be crystal clear.

I remember when Idontcare used to post here before he was run off by a bunch of unsavory posters. I miss his posts

 

[Gikaseixas]

IDC was challenged to justify some of his claims, he didn't like it and run away. Nobody forces you to stop typing. This is the internet brother In the end i do miss his imputs, a very valuable member for the community indeed.

 

[coercitiv]

IDC was challenged to justify some of his claims, he didn't like it and run away.

If by challenge you mean throwing poop at him then by running away you probably mean finding better ways to spend his free time than justifying himself to frustrated anonymous posters. I have no history with the man, nor did I agree with him on a regular basis, but the cowards are elsewhere to be found than on his chair.

Look at Arachnotronic, recently he started making claims of his own, based on whatever sources he can muster. It won't be long until some people will "challenge" him as well (afaik it already started), and although I feel he's more resilient than IDC, and although I have a hard time believing his claims, I still think he should be encouraged to share them properly.

Besides the personal attacks, it was funny to witness the number of false claims IDC had to account for once he made his first slip: everything but the weather.

I have no idea why anybody listens to you, you clearly are out of the industry. Anybody who can still call intel 14nm a decent product or even on par with samsung 14nm in performance can't have worked that recently.

 

[Phynaz]

Somebody here (not me) has IDC's email. Maybe that person could email him pointing out that he is missed.

 

[ShintaiDK]

If by challenge you mean throwing poop at him then by running away you probably mean finding better ways to spend his free time than justifying himself to frustrated anonymous posters. I have no history with the man, nor did I agree with him on a regular basis, but the cowards are elsewhere to be found than on his chair.

Look at Arachnotronic, recently he started making claims of his own, based on whatever sources he can muster. It won't be long until some people will "challenge" him as well (afaik it already started), and although I feel he's more resilient than IDC, and although I have a hard time believing his claims, I still think he should be encouraged to share them properly.

Besides the personal attacks, it was funny to witness the number of false claims IDC had to account for once he made his first slip: everything but the weather.

+1 :thumbsup:

 

[stingerman]

This discussion on process nodes is causing the group to miss the forest for the trees. 2016 will be the year of the SOC for reals. An SOC with heterogenous nodes on the same substrate, get it?

 

[Lepton87]

This discussion on process nodes is causing the group to miss the forest for the trees. 2016 will be the year of the SOC for reals. An SOC with heterogenous nodes on the same substrate, get it?

Like Clarksdale? 32nm core + 45nm graphics portion? That chip was horrible because it effectively didn't have an IMC for the CPU portion not to mention abysmal GPU performance the only saving grace was OC you could get it north of 4.8GHz and be effectively on par with the next gen i3 which was very impressive for both. We don't get that kind of massive improvements from generation to generation with CPU cores and i3 was a special case because it differed significantly from other Nehalem/Westmere products hence why i3 SB was such a huge leap forward.


Clarksdale i3 to SB i3 entry level

http://anandtech.com/bench/product/118?vs=289

ST: improvement in Cinebench 31.51%
MT: 20.5%

WinRAR 3.8 Compression - 300MB Archive

The effect of IMC:

WinRAR 3.8 Compression - 300MB Archive
Time in Seconds - Lower is Better

i3 530 137
i3 2100 86.5

Improvement: 58%


Games:

SC2

Very popular CPU limited game at the time, I can't really explain what's going on

i3 530: 30.7fps
i3 2100: 45.4

improvement: 47%

And yet an overclocked i3 530 could easily keep up with the new SB i3 and in many cases far surpassing it.

 

[stingerman]

Like Clarksdale? 32nm core + 45nm graphics portion? That chip was horrible because it effectively didn't have an IMC for the CPU portion not to mention abysmal GPU performance the only saving grace was OC you could get it north of 4.8GHz and be effectively on par with the next gen i3 which was very impressive for both.

I believe Clarksdale was two different dies within the same package but how the dies connected was not as sophisticated. TSMC is using a Silicon Interposer layer and TSVs to create higher performance/lower power connections between the dies. For example, we might see an A10 that includes 2nd generation high bandwidth memory (HBM) and a Qualcomm die all on the same Interposer. That would significantly improve performance, while significantly reducing power usage and reducing the logic board size.

 

[stingerman]

Theoretically, taken to an extreme, we can imagine all the major dies heterogenously combined in the same chip package, all on the same Silicon Interposer. The Silicon Interposer replacing the function of the logic board substrate. This is a likely goal considering the needs of future Apple Watches and iPhones. Would make for thinner and lighter Watches, iPhones and iPads while allowing for a leap in performance.

On a limb, Would also allow Apple to replace the Intel processor in the Mac.

 

[Eug]

So, is the thread saying:

1) Samsung 10 nm is a bit more dense than Intel 14 nm but not by that much.
2) TSMC 10 nm is more dense than Samsung 10 nm, but not as dense as Intel 10 nm.
3) TSMC 10 nm SoCs likely won't be out in 2016, at least not in any major volume.

Sorry, I'm a n00b.

The reason I ask is because every second article about the Apple A10/A10X is saying it will be on TSMC 10 nm. However, if you read the other tech articles, they would suggest it would be TSMC 16 nm. I guess the one way both claims might make some sense is that they built A10 for either 10 nm or 16 nm, with the intent of TSMC 10 nm, but 10 nm isn't quite there yet so it will be on 16 nm.

However, other articles state that products on TSMC 10 nm are taping out now, and even today's long article in EE Times is suggesting A10 on TSMC 10 nm.

 

[Glo.]

eug,
http://forums.macrumors.com/threads/2016-nmp.1952250/page-38#post-22675233

Timing of posts on both forums is perfect .

 

[Eug]

That wasn't me. I don't have an account there, and I was unaware of that post. Pure coincidence. Well, not really cuz I guess both of us read the same article that just came out today.

 

[Glo.]

Jung would call it "Divine Synchronicity" . At least, I hope it answers your question. A11 is supposed to be on 10 nm process.

 

[Eug]

Well, it's that exact article that specifically suggests A10 is on 10 nm.

Yes there's this in that post you linked:

The foundry taped out last month a 16FF+ device merging a CPU and two HBM2 memory stacks on a silicon interposer to pave the way for its 7nm offering.

But then there's this later on in the same article:

Apple is rumoured to use InFO for 10nm A-series processors in the iPhone 7. Liu said the 10nm InFO process will be ready by June. The 15 x 15mm packages can pack three chips using three redistribution layers of 10 microns each.

 

[Glo.]

http://www.macrumors.com/2016/03/02/tsmc-double-16nm-chip-production/

 

[sm625]

The problem is that intel will have 10nm by next year, while, if history is any indication, TSMC wont have 10nm until 2020.

 

[Arachnotronic]

Well, it's that exact article that specifically suggests A10 is on 10 nm.

Yes there's this in that post you linked:



But then there's this later on in the same article:

A10 is 16FF+, not 10nm.

 

[avAT]

The (old) rumor is that A10 is 16FF+ and A10X is 10nm. Does that help make it more confusing?

Could still be true if A10X is for 1H 2017 this time.

 

[jpiniero]

So, is the thread saying:

1) Samsung 10 nm is a bit more dense than Intel 14 nm but not by that much.
2) TSMC 10 nm is more dense than Samsung 10 nm, but not as dense as Intel 10 nm.
3) TSMC 10 nm SoCs likely won't be out in 2016, at least not in any major volume.

Still seems like 10 nm is just a minor density improvement for both foundries. What they are calling 7 nm now is the one that is comparable to Intel's 14 nm in terms of density.

 

[raghu78]

The (old) rumor is that A10 is 16FF+ and A10X is 10nm. Does that help make it more confusing?

Could still be true if A10X is for 1H 2017 this time.

yeah thats what the rumours are. A9X is already a pretty big chip for a SoC at 147 sq mm. TSMC has confirmed 10nm production in early 2017. So A10X is a likely candidate for the first TSMC 10nm high volume chip.

 

[ShintaiDK]

Still seems like 10 nm is just a minor density improvement for both foundries. What they are calling 7 nm now is the one that is comparable to Intel's 14 nm in terms of density.

That's what happens when PR gets involved.