Samsung Announces Mass Production of 14nm LPP

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Glo.

Platinum Member
Apr 25, 2015
2,669
29
136
#26
Why's that? LPE vs FF+ explains the power difference at maximum frequency.
How can be A9 on 14nm LPP if Samsung started mass production of chips on that process now, not few months ago?

Rhetoric question ;)
 

Andrei.

Senior member
Jan 26, 2015
260
6
116
#27
How can be A9 on 14nm LPP if Samsung started mass production of chips on that process now, not few months ago?

Rhetoric question ;)
Asking the wrong person? I've been saying since the beginning that it's most likely LPE and we've had several confirmations from different sources saying LPP is a 2016 node. You can actually see the LPE vs LPP difference in the transistor structure and the fact that Chipworks and TechInsights have been all mum about it further drives the case.
 

Glo.

Platinum Member
Apr 25, 2015
2,669
29
136
#28
You did not felt that I was agreeing with you, and I was adding to what You have said? ;)

Edit:
Is A9X really that high volume?
A9 was produced by TSMC also for that exact reason. They had to buy capacity from TSMC and A9X alone was not enough.

Other thing is that A9X and its performance would not be achievable on Samsung process, but still.
 
Last edited:

Andrei.

Senior member
Jan 26, 2015
260
6
116
#29
You did not felt that I was agreeing with you, and I was adding to what You have said? ;)
With the amount of stubborn replies that I've received from a few individuals on this topic over the last few weeks it's hard to tell things apart.
A9 was produced by TSMC also for that exact reason. They had to buy capacity from TSMC and A9X alone was not enough.
Actually that makes a lot of sense and I never thought about that.
 

Exophase

Diamond Member
Apr 19, 2012
4,440
0
81
#30
Why's that? LPE vs FF+ explains the power difference at maximum frequency.
So do you think that LPP only improves over LPE power consumption for CPU at maximum frequency on A9? Because if there were a 15% performance improvement for its GPU power consumption, for instance (which by all measures seems to have a higher overall maximum power footprint) that would put it significantly ahead of 16FF+.
 

Andrei.

Senior member
Jan 26, 2015
260
6
116
#31
So do you think that LPP only improves over LPE power consumption for CPU at maximum frequency on A9? Because if there were a 15% performance improvement for its GPU power consumption, for instance (which by all measures seems to have a higher overall maximum power footprint) that would put it significantly ahead of 16FF+.
Not "only" but that's where I expect the largest improvement to happen. The figures that get thrown around a lot are always at peak frequency but both processes more than certainly have different scaling curves, that's why I suspect why we don't see a very large difference in GPU power between both A9's and in overall average battery life, but we see a large difference at peak CPU frequencies. This could also mean that LPP wouldn't bring the full 15% for something like the GPU unless the design implementation was really optimized for the process.

This is anecdotal evidence and might not be related to process at all but for example the Kirin 950 struggles to reach as low voltages as the 7420 at low frequencies even though at high frequencies the voltages clearly favour 16FF+.
 
Last edited:

Exophase

Diamond Member
Apr 19, 2012
4,440
0
81
#32
Not "only" but that's where I expect the largest improvement to happen. The figures that get thrown around a lot are always at peak frequency but both processes more than certainly have different scaling curves, that's why I suspect why we don't see a very large difference in GPU power between both A9's and in overall average battery life, but we see a large difference at peak CPU frequencies. This could also mean that LPP wouldn't bring the full 15% for something like the GPU unless the design implementation was really optimized for the process.

This is anecdotal evidence and might not be related to process at all but for example the Kirin 950 struggles to reach as low voltages as the 7420 at low frequencies even though at high frequencies the voltages clearly favour 16FF+.
My impression is that the GPU power consumption actually trends slightly lower on the Samsung A9s, not merely not as bad as the peak CPU power consumption. It'd be nice to see someone do tests that are even a fraction as thorough as your Android ones have been... Would AT do tests if someone donated the iPhone or money for one? :p

Different dynamic vs static curves (or similar metrics) would explain it; Samsung 14LPE could be better optimized for lower power than TSMC 16FF+. But I wouldn't expect LPP to sacrifice anything, so if it closes the gap at higher clocks it should still have some additional advantage at lower clocks, even if it's not as high.
 

erunion

Senior member
Jan 20, 2013
768
0
0
#33
My sources still stand by the fact that A9 was LPP even in light of this PR, FWIW.
.
I don't see anything in this statement that disapproves that. It's obviously talking about exynos and snapdragon.

Obviously Samsung couldn't make a big PR announcement about chips for Apple, but they are still going to make a big deal out of their own chips.
 

Andrei.

Senior member
Jan 26, 2015
260
6
116
#34
My impression is that the GPU power consumption actually trends slightly lower on the Samsung A9s, not merely not as bad as the peak CPU power consumption. It'd be nice to see someone do tests that are even a fraction as thorough as your Android ones have been... Would AT do tests if someone donated the iPhone or money for one? :p

Different dynamic vs static curves (or similar metrics) would explain it; Samsung 14LPE could be better optimized for lower power than TSMC 16FF+. But I wouldn't expect LPP to sacrifice anything, so if it closes the gap at higher clocks it should still have some additional advantage at lower clocks, even if it's not as high.
We'll have to see - LPP on the 820 seems to be able to go very low in voltages but again only having voltages of the 950 on 16FF+ limits any kind of proper analysis, we'd have to wait on LG's new SoC to get a further data-point.

I'd be happy to get Apple devices in any way or form, of course it won't be as detailed in information due to iOS limitations but I have ideas on how to get some good results out of the phones.
 

raghu78

Diamond Member
Aug 23, 2012
3,999
34
136
#35
Why's that? LPE vs FF+ explains the power difference at maximum frequency.

Don't put any weight onto the 8890 numbers, SLSI has a history of misleading SoC figures (both in terms of overselling (5433) but also vastly underselling (7420) them). That 10% efficiency figure for example is certainly a PR miscommunication as their financial call presentation actually said it was 10% lower power (which together with the perf boost is 44% more efficiency).

Is A9X really that high volume?
I was talking about die size and performance. A9X is the largest mobile SoC right now in the market at 147 sq mm. A9X has a dual core CPU. Each CPU core is a custom ARMv8-A core codenamed Twister with IPC and single thread performance typically 2x of the ARM A57. A9 (TSMC) die size is 104 sq mm and A9(Samsung) is 96 sq mm. Samsung Exynos 7420 is very small at 78 sq mm.

http://www.anandtech.com/show/9824/more-on-apples-a9x-soc

https://www.chipworks.com/about-chipworks/overview/blog/a9-is-tsmc-16nm-finfet-and-samsung-fabbed

http://www.anandtech.com/show/9330/exynos-7420-deep-dive

A9 was produced by TSMC also for that exact reason. They had to buy capacity from TSMC and A9X alone was not enough.

Other thing is that A9X and its performance would not be achievable on Samsung process, but still.
yeah clocking those monster Twister cores at 2.2 Ghz would not be possible on the inferior 14LPE process. 14LPP looks every bit as capable as 16FF+, though we need actual products to get a comparison from chipworks which does detailed transistor performance/power curves.
 
Mar 10, 2006
11,719
121
126
#36
Back in the real world, TSMC's Chairman had the following to say on last night's earnings call:

We estimate our foundry market segment share of 16, 14-nanometer node increases from about 40% in 2015 to above 70% in 2016 exceeding the previous prediction we made in mid-2014.

http://seekingalpha.com/article/381...-results-earnings-call-transcript?part=single

raghu78, I think your prediction that 14LPP will eat the world will be about as good as some of your other predictions on these forums.
 

raghu78

Diamond Member
Aug 23, 2012
3,999
34
136
#37
Back in the real world, TSMC's Chairman had the following to say on last night's earnings call:

http://seekingalpha.com/article/381...-results-earnings-call-transcript?part=single

raghu78, I think your prediction that 14LPP will eat the world will be about as good as some of your other predictions on these forums.
I did not make any prediction on 14LPP but I think it will be quite competitive with TSMC 16FF+. As always chipworks will have transistor performance characteristics reports once 14LPP products are out. As for TSMC saying that they will have 70% of`16/14nm FINFET market share in 2016 I would wait till we see the actual results.

TSMC has guided for Q1 2016 revenue to be in the range of NTD 198-201 billion dollars. Q1 2015 revenue was NTD 222 billion dollars. I don't see that as any indication of a much better year (atleast not a great start to the year).

http://www.tsmc.com/uploadfile/ir/quarterly/2015/16cBq/E/FS.pdf

Snapdragon 820 and Exynos 8890 will be powering a huge chunk of the premium Android smartphone market. We still have to wait and see how Apple sources A10/A10X. We will wait for the year to play itself out in the marketplace. :D
 
Mar 10, 2006
11,719
121
126
#38
I did not make any prediction on 14LPP but I think it will be quite competitive with TSMC 16FF+. As always chipworks will have transistor performance characteristics reports once 14LPP products are out. As for TSMC saying that they will have 70% of`16/14nm FINFET market share in 2016 I would wait till we see the actual results.

TSMC has guided for Q1 2016 revenue to be in the range of NTD 198-201 billion dollars. Q1 2015 revenue was NTD 222 billion dollars. I don't see that as any indication of a much better year (atleast not a great start to the year).

http://www.tsmc.com/uploadfile/ir/quarterly/2015/16cBq/E/FS.pdf

Snapdragon 820 and Exynos 8890 will be powering a huge chunk of the premium Android smartphone market. We still have to wait and see how Apple sources A10/A10X. We will wait for the year to play itself out in the marketplace. :D
A10 is believed to be 100% TSMC, and the company's full-year market share guidance appears to affirm this.
 

raghu78

Diamond Member
Aug 23, 2012
3,999
34
136
#39
A10 is believed to be 100% TSMC, and the company's full-year market share guidance appears to affirm this.
I am waiting to see the year play itself out. TSMC has never faced such competition at previous nodes. I still believe TSMC will be No.1 but I am not sure they will have 70% of FINFET market share. This contest is a great boon for fabless industry. The foundry race to 10nm and 7nm is going to help fabless customers such as Apple, Qualcomm, Nvidia, AMD etc.
 
Feb 6, 2011
1,787
107
136
#40
how much capacity do TMSC and samsung have relative to each other? You know the two sweetest words in the English language........
 
Mar 10, 2006
11,719
121
126
#41
I am waiting to see the year play itself out. TSMC has never faced such competition at previous nodes. I still believe TSMC will be No.1 but I am not sure they will have 70% of FINFET market share. This contest is a great boon for fabless industry. The foundry race to 10nm and 7nm is going to help fabless customers such as Apple, Qualcomm, Nvidia, AMD etc.
I think they will. This stuff is generally known well in advance. Remember that TSMC was pretty blunt in telling us that in 2015 it would see lower 14/16nm share compared to Samsung, and this was way back in 2014.

TSMC management has gained a lot of credibility over the years with successful ramps of 28nm, 20nm and 16nm, so I generally believe what they say these days. They have come a very long way from where they once were.
 

dark zero

Platinum Member
Jun 2, 2015
2,510
8
61
#42
Seems that AMD is about to inaugurate with Zen the Samsung 14 LPP process
 
Mar 13, 2006
10,126
121
126
#44
I am waiting to see the year play itself out. TSMC has never faced such competition at previous nodes. I still believe TSMC will be No.1 but I am not sure they will have 70% of FINFET market share. This contest is a great boon for fabless industry. The foundry race to 10nm and 7nm is going to help fabless customers such as Apple, Qualcomm, Nvidia, AMD etc.
TSMC faced competition greater than this from UMC back in the day. It all came down to capital investment. TSMC won that battle.
 

el etro

Golden Member
Jul 21, 2013
1,581
0
81
#45
Fottemberg was right. Samsung delivered first the 1st generation finfets, TSMC delivered first the second generation finfets.
 

prtskg

Senior member
Oct 26, 2015
209
10
71
#46
Seems that AMD is about to inaugurate with Zen the Samsung 14 LPP process
No. Samsung's Exynos 8890 and Qualcomm's Snapdragon 820 SoCs will be first to use 14nm LPP. English.etnews.com reported that AMD will use Samsung's foundry only after Q2 2016. This report is from INDUSTRY SOURCES so consider it as rumor.

Fottemberg was right. Samsung delivered first the 1st generation finfets, TSMC delivered first the second generation finfets.
Do we have any announcements from TSMC about achieving volume production in 16nm ff+? I know some people have 'industry sources' claiming so but I'd rather believe company's announcements.
 
Apr 12, 2015
107
0
36
#47
Only if you are taking the average with AMD's being ready in 3 months while Nvidia's are ready in 9.

Trolling is not allowed here
Markfw900
Claiming AMD has an advantage over intel or Nvidia is considered trolling on these boards, please think before you post.

Mod baiting is also not allowed here.
Administrator allisolm
 
Last edited by a moderator:

krumme

Diamond Member
Oct 9, 2009
5,734
100
136
#48
how much capacity do TMSC and samsung have relative to each other? You know the two sweetest words in the English language........
And how much does older than 20nm nodes contribute to profit?

And how much profit is there in selling phones instead of wafers?
 

Similar threads



ASK THE COMMUNITY

TRENDING THREADS