Throw in 15% from process and a few percent from tweaked internals and the A10 looks like an A8; about 25% faster, 15% of which comes from frequency. So A10 at maybe 2.2GHz, A10X at maybe 2.6 GHz. That's pretty much what I expected, and this is about as close to confirmation as I think we'll ever get. The real question, then remains what it was: whether TSMC and/or Samsung get to 10nm in time for the A11... (which suggests they get there before Intel --- cue the complaints about what does/does not count as "real" 10nm...)
GlobalFoundries is fabbing the low-end Polaris chips using Samsung's 14nm design (since they couldn't get their own right). TSMC's 16nm will still be used for the larger and more complex chips.
AMD clearly states that Polaris is using 14nm FinFET confidently even in the Title. So that means TSMC is out for Polaris.
Since it is mid 2016, assuming that GloFo, which has licensed Samsung's 14nm FinFET technology, is setting up LPP lines for that, which means that it might be in LPP.
Looking at the timelines for Zen, it looks highly to be 14nm LPP rather than 14nm LPE if it were to be fabbed at GF or SMSNG.
I could be wrong, but I think GF went straight to LPP, and totally skipped LPE. Future AMD CPU/GPU's will definitely be at least LPP, or possibly a future revision even.
No. LPE entered volume production in GF in Q2. http://fudzilla.com/news/processors/37647-globalfo... LPP was to start volume production in December or in early 2016. There isn't any news that it has started volume production yet.
Why would AMD be using "low power" anything for their desktop GPUs and CPUs? The foundry announcement that would be more pertinent would be one mentioning "high performance" or the like. Anyway, hasn't AMD already indicated that Polaris would be out "for the back to school season"? I don't know the context of that statement but if it was a neutral (non-leading) context then that seems to suggest July-August.
From http://www.samsung.com/semiconductor/foundry/proce... "14LPP (Performance boosted edition) is the 2nd FinFET generation which the performance is enhanced up to 10%. 14LPP is the single platform for every application designs with the improved performance for computing/Network designs and the lowered power consumption for Mobile/Consumer designs. 14LPP will be the main process technology offering in 2016 and after."
Samsung now calls LPE 'early edition' and LPP 'performance boosted edition'. Officially low power early and low power plus aren't used as it causes confusion to some people that these nodes aren't geared for high performance chips. So I think tech sites should update themselves too. http://www.samsung.com/semiconductor/foundry/proce...
LPE is already fairly competitive with FF+, and LPP is way better than LPE. So what makes you think TSMC would still hold the edge? They need FFC to compete with LPP, who knows when FFC will be ready for volume.
So at this point, has Samsung caught Intel or is Intel's 14nm better? Remember everyone talking about how Intel had a 2 year lead or a process and a half lead. Seems not to be the case anymore.
I don't think it's that simple. There are a bunch of factors determining the quality of the node. I don't think there's any single number distillate that will accurately describe it.
TSMCs process is 20nm based, which us why it's labeled as 16nm (20nm with FinFet and other enhancements make it, at least according to TSMC, equivalent to 16nm)
Samsung's (and licensee GF's) process is (apparently) true 14nm.
I think you are correct. It was mentioned in TSMC's earnings call last year that 16 FinFET+ out performs 14LPP. Also, it explains why TSMC will have 70% of the FinFET market this year.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
32 Comments
Back to Article
lilmoe - Thursday, January 14, 2016 - link
If you don't mind me asking, which process was Apple's A9 based on? I already thought it was 14nm LPP?hlovatt - Thursday, January 14, 2016 - link
According to the article the existing process is called 14LPE (Low-Power Early)lilmoe - Thursday, January 14, 2016 - link
Well yeah, but I could swear 14nm LPP was mentioned in the A9's article...revanchrist - Thursday, January 14, 2016 - link
I could swear 7nm FunFiT was mentioned too.prtskg - Thursday, January 14, 2016 - link
While many people thought it was lpp, A9 used lpe. Since lpp has just entered volume production, it couldn't be used in A9.name99 - Thursday, January 14, 2016 - link
Throw in 15% from process and a few percent from tweaked internals and the A10 looks like an A8; about 25% faster, 15% of which comes from frequency.So A10 at maybe 2.2GHz, A10X at maybe 2.6 GHz.
That's pretty much what I expected, and this is about as close to confirmation as I think we'll ever get.
The real question, then remains what it was: whether TSMC and/or Samsung get to 10nm in time for the A11... (which suggests they get there before Intel --- cue the complaints about what does/does not count as "real" 10nm...)
khon - Thursday, January 14, 2016 - link
Good news, hopefully this means new Polaris GPUs before long, the move to 14nm is long overdue.Shadow7037932 - Thursday, January 14, 2016 - link
Is Samsung doing the AMD GPUs? I thought TSMC was doing that.The_Assimilator - Thursday, January 14, 2016 - link
GlobalFoundries is fabbing the low-end Polaris chips using Samsung's 14nm design (since they couldn't get their own right). TSMC's 16nm will still be used for the larger and more complex chips.mdriftmeyer - Thursday, January 14, 2016 - link
Wrong. AMD is using Samsung and Global Foundries across the board. TSMC is out. Nvidia is using TSMC exclusively.extide - Thursday, January 14, 2016 - link
This is what was originally reported, but a few places are now confirming that it will be 14nm only, and pretty much it will all be GF.rocketbuddha - Thursday, January 14, 2016 - link
http://www.amd.com/en-us/press-releases/Pages/amd-...AMD clearly states that Polaris is using 14nm FinFET confidently even in the Title. So that means TSMC is out for Polaris.
Since it is mid 2016, assuming that GloFo, which has licensed Samsung's 14nm FinFET technology, is setting up LPP lines for that, which means that it might be in LPP.
Looking at the timelines for Zen, it looks highly to be 14nm LPP rather than 14nm LPE if it were to be fabbed at GF or SMSNG.
extide - Thursday, January 14, 2016 - link
I could be wrong, but I think GF went straight to LPP, and totally skipped LPE. Future AMD CPU/GPU's will definitely be at least LPP, or possibly a future revision even.prtskg - Thursday, January 14, 2016 - link
No. LPE entered volume production in GF in Q2.http://fudzilla.com/news/processors/37647-globalfo...
LPP was to start volume production in December or in early 2016. There isn't any news that it has started volume production yet.
Yojimbo - Sunday, January 17, 2016 - link
Why would AMD be using "low power" anything for their desktop GPUs and CPUs? The foundry announcement that would be more pertinent would be one mentioning "high performance" or the like. Anyway, hasn't AMD already indicated that Polaris would be out "for the back to school season"? I don't know the context of that statement but if it was a neutral (non-leading) context then that seems to suggest July-August.Yojimbo - Sunday, January 17, 2016 - link
Nevermind, I was wrong.From http://www.samsung.com/semiconductor/foundry/proce...
"14LPP (Performance boosted edition) is the 2nd FinFET generation which the performance is enhanced up to 10%. 14LPP is the single platform for every application designs with the improved performance for computing/Network designs and the lowered power consumption for Mobile/Consumer designs. 14LPP will be the main process technology offering in 2016 and after."
prtskg - Thursday, January 14, 2016 - link
Samsung now calls LPE 'early edition' and LPP 'performance boosted edition'. Officially low power early and low power plus aren't used as it causes confusion to some people that these nodes aren't geared for high performance chips. So I think tech sites should update themselves too.http://www.samsung.com/semiconductor/foundry/proce...
zodiacfml - Thursday, January 14, 2016 - link
What people? I don't know how this would matter to some people unless the change is targeted to investors.prtskg - Friday, January 15, 2016 - link
People who invest, people who follow tech, everyone. Why should we use an 'old' term when it isn't a good one?Yojimbo - Sunday, January 17, 2016 - link
Ahh, thanksiwod - Thursday, January 14, 2016 - link
I think, this will edge out TSMC 16nm FF+ slightly. But TSMC 16nm FFC is coming out this year as well.kaze102477 - Friday, January 15, 2016 - link
No,TSMC 16+ still edge out 14nm LPP.Alexvrb - Sunday, January 17, 2016 - link
LPE is already fairly competitive with FF+, and LPP is way better than LPE. So what makes you think TSMC would still hold the edge? They need FFC to compete with LPP, who knows when FFC will be ready for volume.olderkid - Thursday, January 14, 2016 - link
So at this point, has Samsung caught Intel or is Intel's 14nm better? Remember everyone talking about how Intel had a 2 year lead or a process and a half lead. Seems not to be the case anymore.extide - Thursday, January 14, 2016 - link
Intel's 14nm is better. It's a true shrink from 22nm.Hulk - Thursday, January 14, 2016 - link
So in "Intel terms" what size is this process? 16nm? 18nm? Apples-to-Apples.Yojimbo - Sunday, January 17, 2016 - link
I don't think it's that simple. There are a bunch of factors determining the quality of the node. I don't think there's any single number distillate that will accurately describe it.T1beriu - Thursday, January 14, 2016 - link
I could be wrong, but If I'm remembering correctly, TSMC and Samsung added FinFet to their 20nm and calling it 14/16.eldakka - Thursday, January 14, 2016 - link
TSMCs process is 20nm based, which us why it's labeled as 16nm (20nm with FinFet and other enhancements make it, at least according to TSMC, equivalent to 16nm)Samsung's (and licensee GF's) process is (apparently) true 14nm.
kaze102477 - Friday, January 15, 2016 - link
But TSMC's 16+ still better than SS's 14LPPlefty2 - Sunday, January 17, 2016 - link
I think you are correct. It was mentioned in TSMC's earnings call last year that 16 FinFET+ out performs 14LPP. Also, it explains why TSMC will have 70% of the FinFET market this year.rocketbuddha - Saturday, January 16, 2016 - link
Nope both are the lower nanometer FEOL with 20nm BEOL.FEOL - Front End Of Line
BEOL - Back End of Line