Samsung and GLOBALFOUNDRIES Forge Strategic Collaboration to Deliver 14nm FinFET

[blackened23]

why do you think anybody should believe you ? whats your credibility when compared against anandtech? sorry I don't care about what you have to say. Cherrytrail against A8. Lets see how that plays out before calling cherrytrail a game changer.

Because Cherry trail and the A8 will absolutely be competing for a spot in the same Android and Windows devices with OEMs such as Dell, HP, HTC, LG, Xiomi, right? I can't believe how absurd this argument is getting.

Oh hey. I bet the A8 will perform better than the Cherry Trail in iOS. WHAT A GAME CHANGER! An OS that CANNOT be run by any other SOC runs best with the A8. Holy crap, hold the press. This is truly game changing. Cherry Trail performs worse in iOS because hell, IT CANT EVEN RUN IOS! Or let's just flip that around, I bet the A8 will perform AMAZINGLY well in windows and android. Oh wait.

 

[jdubs03]

why do you think anybody should believe you ? whats your credibility when compared against anandtech? sorry I don't care about what you have to say. Cherrytrail against A8. Lets see how that plays out before calling cherrytrail a game changer.

Lol, you deflect my points that counter yours, and then state I have no credibility, even though I am probably one of the least biased posters on this forum. You're out of line. I'm not aiming to usurp Anand or any other employees credit, I'm just reporting the results of my own machine, and that was at a time where there wasn't many benchmark postings so I wanted to help the community. Why do you think there are other threads on here asking for benchmarks for Cinebench or Povray?

You're so incredibly biased it's not even funny. And it shows, because you can't even acknowledge that CT's GPU will be massively upgraded. I'll throw my un-credible prediction out there, CT will perform within at least 85% of HD4400, and that is the safe guess.

 

[raghu78]

Because Cherry trail and the A8 will absolutely be competing for a spot in the same Android and Windows devices with OEMs such as Dell, HP, HTC, LG, Xiomi, right? I can't believe how absurd this argument is getting.

Oh hey. I bet the A8 will perform better than the Cherry Trail in iOS. WHAT A GAME CHANGER! An OS that CANNOT be run by any other SOC runs best with the A8. Holy crap, hold the press. This is truly game changing. Cherry Trail performs worse in iOS because hell, IT CANT EVEN RUN IOS! Or let's just flip that around, I bet the A8 will perform AMAZINGLY well in windows and android. Oh wait.

these upcoming Android flagship phones are going to be benchmarked against the A8 running iphone 6. The best part is Apple is going to go after precisely these companies (Samsung, HTC, LG, Huwaei , Xaomi . btw who cares about Dell and HP in phones) and their large phone flagships with the iphone 6 5.5 inch. have an answer for that.

 

[raghu78]

Lol, you deflect my points that counter yours, and then state I have no credibility, even though I am probably one of the least biased posters on this forum. You're out of line. I'm not aiming to usurp Anand or any other employees credit, I'm just reporting the results of my own machine, and that was at a time where there wasn't many benchmark postings so I wanted to help the community. Why do you think there are other threads on here asking for benchmarks for Cinebench or Povray?

You're so incredibly biased it's not even funny. And it shows, because you can't even acknowledge that CT's GPU will be massively upgraded.

for sure post your comparisons. do it. but for anyone to take your ipad air vs t100 comparison seriously you need to probably post a youtube video of the testing with screenshots.

oh btw Intel's products don't exist in a vacuum. Cherrytrail's GPU might be upgraded. so what. cherrytrail will be compared against Apple A8 and Snapdragon 810 and then only can you really rate their effort.

 

[blackened23]

these upcoming Android flagship phones are going to be benchmarked against the A8 running iphone 6. The best part is Apple is going to go after precisely these companies (Samsung, HTC, LG, Huwaei , Xaomi . btw who cares about Dell and HP in phones) and their large phone flagships with the iphone 6 5.5 inch. have an answer for that.

I don't dislike Apple products. I actually like them. But there are plenty of people who buy non Apple products, and Android is a very healthy ecosystem with some very high selling devices. Let's not pretend that Apple is the only game in town. They're huge, yes, but especially among phones there is very healthy competition there. Let's not kid ourselves here. You're grasping for straws to bad-mouth intel and the A7/A8 is your best ammo. Even THOUGH those SOCs have absolutely nothing to do with Windows or Android, and are not competing with intel on an SOC level. Your comparisons are absurd. But since you so thoroughly want to badmouth intel, it's a convenient source of ammunition. Whatever man. You do your thing. You know that Android premium phones and tablets sell just fine, and intel owns the high end performance area with ultrabooks/convertibles. Intel may or may not get appreciable design wins with Cherry Trail, but comparing it against an SOC that it will never directly compete against for OEM wins is so silly it is truly mind bogging. But it's your best ammo, right?

 

[jdubs03]

for sure post your comparisons. do it. but for anyone to take your ipad air vs t100 comparison seriously you need to probably post a youtube video of the testing with screenshots.

Nope, I'm not going to waste my time to prove a silly point, I posted my results back in November, I don't have to repost or use extra effort. I also ran the results multiple times to ensure a solid average.

The GPU performance of CT is going definitely go up against the A8, and the S810 whenever it comes out. But with Gen8 being a significant improvement over the Haswell era EU's, and having 80% of the EU's of HD4400, its pretty hard to see CT not performing near or above HD4400. The GX6650 in the A8 will be around Tegra K1 levels, which is significantly below HD4400 performance. The S805 right now is getting whipped by TK1, and that is the top-end solution til likely Q2 of next year -> S810. By then Erista (TM1?) will be near, and Broxton (Gen 9) may not be too far off.

 

[raghu78]

I don't dislike Apple products. I actually like them. But there are plenty of people who buy non Apple products, and Android is a very healthy ecosystem with some very high selling devices. Let's not pretend that Apple is the only game in town. They're huge, yes, but especially among phones there is very healthy competition there. Let's not kid ourselves here. You're grasping for straws to bad-mouth intel and the A7/A8 is your best ammo. Even THOUGH those SOCs have absolutely nothing to do with Windows or Android, and are not competing with intel on an SOC level. Your comparisons are absurd. But since you so thoroughly want to badmouth intel, it's a convenient source of ammunition. Whatever man. You do your thing. You know that Android premium phones and tablets sell just fine, and intel owns the high end performance area with ultrabooks/convertibles. Intel may or may not get appreciable design wins with Cherry Trail, but comparing it against an SOC that it will never directly compete against for OEM wins is so silly it is truly mind bogging. But it's your best ammo, right?

I don't care about Intel. I am just not a believer in "Intel will dominate every market they enter" kind of theme

http://www.canalys.com/newsroom/third-smart-phones-shipped-q1-had-5-plus-displays

"Smart phones with 5" and larger screens grew 369% - a substantially greater rate of growth than the overall market. Worldwide, they represented just over a third of shipments (34%), and in Greater China the figure hit 39%, and 43% in Asia Pacific.

This is still a market segment led by Samsung, but the trend is unmistakably toward larger-screen handsets at the high end of the market. It held a 44% share of devices with displays of 5" and above, and 53% if the view is narrowed to look at 5.5"-plus displays,’ said Canalys Analyst Jessica Kwee.

‘But many other vendors, such as Lenovo, Huawei, LG and Sony, have also achieved significant volumes in this space with products at the top end of their portfolios. Consumers now expect high-end devices to have large displays, and Apple’s absence in this market will clearly not last long.

It is notable that 5" and above displays featured on almost half (47%) of smart phones with an unlocked retail price of US$500 or more. Of the remaining 53% of high-end smart phones, 87% were iPhones. Apple plainly needs a larger-screen smart phone to remain competitive, and it will look to address this in the coming months.’"

So the way it is Apple dominates the segment where it has a product and is missing out massively on one of the fastest growing portions (and most likely profitable portions) of the smartphone market which is the 5+ inch smartphone market. Apple is addressing the situation with the 5.5 inch iphone 6 (albeit a bit later than they should have done).

 

[raghu78]

Nope, I'm not going to waste my time to prove a silly point, I posted my results back in November, I don't have to repost or use extra effort. I also ran the results multiple times to ensure a solid average.

then I don't care about what you have to say.

 

[jdubs03]

I don't dislike Apple products. I actually like them. But there are plenty of people who buy non Apple products, and Android is a very healthy ecosystem with some very high selling devices. Let's not pretend that Apple is the only game in town. They're huge, yes, but especially among phones there is very healthy competition there. Let's not kid ourselves here. You're grasping for straws to bad-mouth intel and the A7/A8 is your best ammo. Even THOUGH those SOCs have absolutely nothing to do with Windows or Android, and are not competing with intel on an SOC level. Your comparisons are absurd. But since you so thoroughly want to badmouth intel, it's a convenient source of ammunition. Whatever man. You do your thing. You know that Android premium phones and tablets sell just fine, and intel owns the high end performance area with ultrabooks/convertibles. Intel may or may not get appreciable design wins with Cherry Trail, but comparing it against an SOC that it will never directly compete against for OEM wins is so silly it is truly mind bogging. But it's your best ammo, right?

Right. Incredulous arguments, and refusal to take information as is. But hey if it were pro-apple there would be no need for "proof". I just hope Intel and Google can get their binary issues solved, because that seems to be the most limiting factor with app compatibility.

 

[Fjodor2001]

Samsungs 14nm looks to be alittle worse than TSMC 16nm.

Can you please show us a list of metrics where Samsung/GF 14 nm is compared to TSMC 16 nm to back that statement up?

Since its you claiming that TSMC and Samsung is gaining on Intel. Let me ask you if you think that a 14% density improvement from 20nm to 14nm in the case of Samsung justifies calling it 14nm. Or TSMCs 15% from 20nm to 16nm. Not to mention all the design tools for both nodes are based on 20nm.

For Samsung the metal pitch is the same 64nm and the poly pitch is reduced from 90nm to 80nm for 14nm FF vs 20nm planar.

Or as said by someone else:

Can you please answer the question instead? If not, I take it you have no solid evidence to back up your statement us usual.

 

[witeken]

then I don't care about what you have to say.

But I do.

 

[raghu78]

The GPU performance of CT is going definitely go up against the A8, and the S810 whenever it comes out. But with Gen8 being a significant improvement over the Haswell era EU's, and having 80% of the EU's of HD4400, its pretty hard to see CT not performing near or above HD4400. The GX6650 in the A8 will be around Tegra K1 levels, which is significantly below HD4400 performance. The S805 right now is getting whipped by TK1, and that is the top-end solution til likely Q2 of next year -> S810. By then Erista (TM1?) will be near, and Broxton (Gen 9) may not be too far off.

all you are doing is guessing GPU performance of Cherrytrail and A8. do you know what GPU clocks will Cherrytrail run at ? Same for GX6650 on a TSMC 20nm process. Do you know the GPU clocks in the A8 ? btw PowerVR is one of the most power efficient mobile GPU architectures. Its probable that Kepler might be more efficient and even on 28nm Tegra K1 could be competitive against a 20nm A8 running GX6650. But we need to wait for final product before judging performance and efficiency.

http://www.anandtech.com/show/7793/imaginations-powervr-rogue-architecture-exposed
http://www.anandtech.com/show/7796/...ounces-new-high-end-mobile-gpu-powervr-gx6650

As for Intel's graphics efficiency they are bottom of the ladder. Without the EDRAM on Iris Pro their graphics chips suck in terms of perf/sq mm and perf/watt. So lets wait and see how cherrytrail fares against A8 and S810.

but the key here is cherrytrail won't be shipping before Q1 2015. same timeline as S810. Devices with these chips will be on shelves in mid 2015. A8 will be available in late Sep 2014.

 

[witeken]

A8 will be launched in Q4, not mid-2015. You say that you have to wait to see performance, and one sentence later you start complaining about Gen7. Gen7 ain't Gen8.

In general terms, Intel’s Ben Widawsky, who works on Intel’s Linux graphics driver efforts, says that “Broadwell graphics bring some of the biggest changes we’ve seen on the execution and memory management side of the GPU… [the changes] dwarf any other silicon iteration during my tenure, and certainly can compete with the likes of the gen3->gen4 changes.”

 

[raghu78]

A8 will be launched in Q4, not mid-2015.

Did I say anything different. I said A8 in late Sep 2014. worst case early Oct 2014.

You say that you have to wait to see performance, and one sentence later you start complaining about Gen7. Gen7 ain't Gen8.

nobody knows how Gen8 performs or how efficient it is. So pardon me if there is a healthy dose of skepticism if someone says that Intel will bring out a highly efficient mobile GPU or brags about Cherrytail GPU performance.

 

[blackened23]

...but we know how the A8 performs. Right? And we know the A8 will run Windows and Android, finally, right? Then the comparisons would be somewhat valid, maybe.

 

[Ajay]

TSMC has literally no competition at 20nm. Also TSMC's yield learning at 20nm helps with 16FF and 16FF+ as they share the same BEOL.

I don't think so. Building FinFets will be completely new to TSMC - so they won't really benefit from 20nm BEOL experience.

 

[witeken]

nobody knows how Gen8 performs or how efficient it is. So pardon me if there is a healthy dose of skepticism if someone says that Intel will bring out a highly efficient mobile GPU or brags about Cherrytail GPU performance.

-4X as many EUs
-new node
-massively updated architecture

You can't beat those specs. Or can you?

 

[raghu78]

I don't think so. Building FinFets will be completely new to TSMC - so they won't really benefit from 20nm BEOL experience.

http://www.tsmc.com/uploadfile/ir/quarterly/2014/2YN1J/E/TSMC%202Q14%20transcript.pdf

"The 16-nanometer development leverages off 20-SoC learning and is moving forward smoothly. Our 16-nanometer is more than competitive,combining performance, density and yields considerations. 16-nanometer applications cover a wide range including baseband, application processors, consumer SoCs, GPU, network processors, hard disk drive, FPGA, servers and CPUs. Volume production of 16-nanometer is expectedto begin in late 2015 and there will be a fast ramp up in 2016. The ecosystem for 16-nanometer designs is current and ready.A few years ago, in order to take advantage of special market opportunities, we chose to develop 20-SoC first and then quickly follow with16-nanometer. We chose this sequence to maximize our market share in the 20/16-nanometer generation. As the 20/16 foundry competition unfolds, we believe our decision to have been correct."

Samsung might get to FINFET first but in the long run what matters is performance, density and the big elephant in the room called YIELDS. If Intel is having yield problems at 14nm FINFET which brings dual patterning immersion litho with FINFET for the first time then suffice to say foundries are not going to have it easy either. TSMC has taken a more measured approach to reduce their risks. Samsung seems to have given up on 20nm to speeden up 14 FINFET. But the question is how good are their yields going to be ? Nobody knows that.

 

[NTMBK]

-4X as many EUs
-new node
-massively updated architecture

You can't beat those specs. Or can you?

"Massively updated" isn't necessarily better. Just look at the Bulldozer debacle!

Intel have completely broken the "Tick Tock" model for GPUs. The whole idea of tick-tock was to reduce risk by changing either process or architecture, but not both- shrink, refine, shrink, refine. Working with a know arch makes dialling in the new process easier, and working with a known process makes refining the architecture easier. But Intel have completely abandoned that for the GPU, and now surprise, surprise, Broadwell is heavily delayed. Maybe Intel should have stuck to Tick Tock...

 

[raghu78]

-4X as many EUs
-new node
-massively updated architecture

You can't beat those specs. Or can you?

Specs don't mean anything. What matters is architecture and efficiency. AMD Piledriver can clock to 5 Ghz but their architecture sucks. so nobody cares about specs. lets see what Intel has in store for Cherrytrail.

 

[raghu78]

"Massively updated" isn't necessarily better. Just look at the Bulldozer debacle!

Intel have completely broken the "Tick Tock" model for GPUs. The whole idea of tick-tock was to reduce risk by changing either process or architecture, but not both- shrink, refine, shrink, refine. Working with a know arch makes dialling in the new process easier, and working with a known process makes refining the architecture easier. But Intel have completely abandoned that for the GPU, and now surprise, surprise, Broadwell is heavily delayed. Maybe Intel should have stuck to Tick Tock...

well said. Also we have to see if Broadwell comes out highly crippled for yield reasons.

http://semiaccurate.com/2014/07/11/intel-castrates-broadwell-gutting-performance/

 

[teejee]

...but we know how the A8 performs. Right? And we know the A8 will run Windows and Android, finally, right? Then the comparisons would be somewhat valid, maybe.

An OS causes very little overhead for normal CPU benchmarks. So CPU benchmarks are in general comparable between OS.

GPU benchmarks is a little bit more difficult to compare between OS though. Different OpenGL stacks could make quite some difference.

 

[Khato]

I don't think so. Building FinFets will be completely new to TSMC - so they won't really benefit from 20nm BEOL experience.

http://www.tsmc.com/uploadfile/ir/quarterly/2014/2YN1J/E/TSMC%202Q14%20transcript.pdf

"The 16-nanometer development leverages off 20-SoC learning and is moving forward smoothly. Our 16-nanometer is more than competitive,combining performance, density and yields considerations. 16-nanometer applications cover a wide range including baseband, application processors, consumer SoCs, GPU, network processors, hard disk drive, FPGA, servers and CPUs. Volume production of 16-nanometer is expectedto begin in late 2015 and there will be a fast ramp up in 2016. The ecosystem for 16-nanometer designs is current and ready.A few years ago, in order to take advantage of special market opportunities, we chose to develop 20-SoC first and then quickly follow with16-nanometer. We chose this sequence to maximize our market share in the 20/16-nanometer generation. As the 20/16 foundry competition unfolds, we believe our decision to have been correct."

Samsung might get to FINFET first but in the long run what matters is performance, density and the big elephant in the room called YIELDS. If Intel is having yield problems at 14nm FINFET which brings dual patterning immersion litho with FINFET for the first time then suffice to say foundries are not going to have it easy either. TSMC has taken a more measured approach to reduce their risks. Samsung seems to have given up on 20nm to speeden up 14 FINFET. But the question is how good are their yields going to be ? Nobody knows that.

And what precisely would you expect TSMC to say? They keep repeating that 'our 20nm experience will make 16nm easy' to justify their decision to not drop their 20nm process for FinFET like Samsung did. What's the truth of the matter though? It's pretty simple - their 20nm experience will make about a third of their '16nm' easy. Because all they get to port without changes are a portion of the BEOL steps, and then of course some of what they learned from 20nm with respect to double patterning will also be applicable. The remainder of the BEOL will need quite a bit of modification to support FinFET rather than planar transistors, and then the FEOL will be completely different. By the way, it's the FEOL that's the real challenge when it comes to bringing FinFETs to mass production thanks to the additional dimension. Oh yeah, and it's the portions of the process which TSMC has to change due to the nature of FinFETs which have the largest effect upon yields by the way.

You're welcome to blindly follow TSMC's attempt to coax investor confidence of course, but I'm certain others will understand why the party line should be questioned.

 

[witeken]

"Massively updated" isn't necessarily better. Just look at the Bulldozer debacle!

Intel isn't AMD...

Intel have completely broken the "Tick Tock" model for GPUs. The whole idea of tick-tock was to reduce risk by changing either process or architecture, but not both- shrink, refine, shrink, refine. Working with a know arch makes dialling in the new process easier, and working with a known process makes refining the architecture easier. But Intel have completely abandoned that for the GPU, and now surprise, surprise, Broadwell is heavily delayed. Maybe Intel should have stuck to Tick Tock...

We don't know what caused those yield problems, but it seems that it's because of the lithography (lack of EUV). I wouldn't be so fast blaming the GPU. Intel should be competent enough not to run into those issues, and as far as I know, 14nm really is an exception. BTW, Intel is still on a Tick-Tock cadence.

 

[raghu78]

And what precisely would you expect TSMC to say? They keep repeating that 'our 20nm experience will make 16nm easy' to justify their decision to not drop their 20nm process for FinFET like Samsung did. What's the truth of the matter though? It's pretty simple - their 20nm experience will make about a third of their '16nm' easy. Because all they get to port without changes are a portion of the BEOL steps, and then of course some of what they learned from 20nm with respect to double patterning will also be applicable. The remainder of the BEOL will need quite a bit of modification to support FinFET rather than planar transistors, and then the FEOL will be completely different. By the way, it's the FEOL that's the real challenge when it comes to bringing FinFETs to mass production thanks to the additional dimension. Oh yeah, and it's the portions of the process which TSMC has to change due to the nature of FinFETs which have the largest effect upon yields by the way.

You're welcome to blindly follow TSMC's attempt to coax investor confidence of course, but I'm certain others will understand why the party line should be questioned.

At these leading edge nodes all company statements have to be taken with a healthy dose of skepticism. Its the same for Intel or TSMC or Samsung. what matters is actual products in the marketplace made using these leading edge process nodes and their availability.