samsung to fab 14nm for amd

illli

[H]ard|Gawd
Joined
Oct 26, 2005
Messages
1,295
that article is confusing. first it says only samsung, then later down the page it says both samsung and global foundries?
 

dragonstongue

2[H]4U
Joined
Nov 18, 2008
Messages
3,161
GF, Samsung, as well as other partners such as IBM had a massive hand in development behind HBM, 14nm, FinFETT and so forth, so it is no surprise in this regard.

Also this is very "early" news anyways, as they had announced this back at least as far as April, 2014
essentially they can work in tandem for 14nm production, as they share similar foundry process and so forth.

I would imagine that as it stands AMD is the only 14nm client for GF, and Samsung is probably for specific models/lines or only for overflow needs? I would imagine, maybe, they found out that Samsung 14nm was a bit better for raw speed not as much for power save, so they slap that for use of example Zen APU, whereas the GF process was better power save at higher speed so they choose to use this for their GPU and high end FX.

who knows :D
http://www.globalfoundries.com/news...ering-of-14nm-finfet-semiconductor-technology
 

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
hmm well I'm going to say it right now, using both Samsung and Global foundries to do the same chip, they will forgo using custom libraries, that gives their competitors an advantage both on GPU (nV has been using custom libraries for their products since the g80) and of course CPU's where Intel has been custom tailoring ever process for their CPU's. And this is why I don't think they will be using both foundries at least for their CPU's.....
 
Last edited:

Yakk

Supreme [H]ardness
Joined
Nov 5, 2010
Messages
5,810
I believe GF & Samsung use the same manufacturing process, so it makes sense if AMD is to use them not use TSMC which would require a complete redesign.

There is not enough volume to warrant doing what Apple did IMO.
 

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
custom libraries are still tailored to the fab and not the process. 2 different fabs using the same process will still have different custom libraries. AMD doesn't create the custom libraries, well they have some input to the Fabs, but the Fab will create them.
 

dragonstongue

2[H]4U
Joined
Nov 18, 2008
Messages
3,161
As I understood, they go matched 14nm process and production, however Samsung and GF do have their own custom "recipes" as far as gate first/gate last, high density libraries, power and drain gating redundancy and all kinds of fancier stuff.

Anyways, last I read on this was they both can produce the 100% exact thing because of how GF/Samsung/IBM drafted the 14nm fabs and process design, not counting any custom variants they also offer, though sharing directly between capable fabs very similar process.

This I understood to mean want to benefit from low power lower performance use process X at GF, for low power high performance use process Y at Samsung and forth but by being able to share the same mask it basically is the exact same maybe some +/- I assume because they are functionally the same setup (granted lot per lot changes even in same factory) this should mean between the 2 or 4 fabs capable of producing at same time, same exact chip, maybe it will allow a much better binning process, as well a very high production capacity, or switch to a "better" process when need be because of these fabs also producing chips for others that maybe they find something that would help drastically)

This would kind of thing reminds me with Intel as an example, controls the design, production and marketing, so they know exactly what they are doing and what they need to do, but also means they can bump against a wall and not as able change to a better recipe as they are pretty much identical so they wouldn't be anything but marginal the same exact chip anyways, however by AMD/"partners" all being available, there does provide that option of being able to essentially go with the best "lot" after x amount production run(once you know the basic yields from test runs) you cannot always be right, you can only go with what testing tells you, maybe one is slightly more aggressive, maybe one is scaled down a bit etc.

I suppose when its concrete that X is produced at Fab Y GF or X is produced at Fab Z Samsung etc, it really is hyperbole.
 
Last edited:

dragonstongue

2[H]4U
Joined
Nov 18, 2008
Messages
3,161
What I hope it means either way is their time to market is on the ball or sooner which is supposed to be the next APU which will be able to use the AM4 board sometime in spring-summer produced GF 28nm, next will be Arctic Islands which will probably be TSMC for 14nm and Zen will probably be a mix of GF and Samsung Fabs for various market segments, as Samsung is obviously the closest ones these days besides Intel for design, process, production, capacity.
 

Derfnofred

Gawd
Joined
Dec 11, 2009
Messages
599
Razor alludes to this, but even on the same process, each fab will be a little bit different.
 

dragonstongue

2[H]4U
Joined
Nov 18, 2008
Messages
3,161
http://hexus.net/business/news/components/89078-samsung-produce-greenland-gpus-amd-starting-q2-2016/

An industry insider told the Korea Economic Times "Because Samsung Electronics and GF have same IP for 14-nano processing, chips that are designed by AMD will all be produced at both factories". The split production should also help if there is unexpectedly large demand.

likely differences will be there, but I doubt it would be nearly as drastic as it would be say 14m, GF vs 14nm Intel etc, I IMHO think it will modest difference such as same exact chip all plants maybe they can better bin the better parts because of the combined testing taking this into account, so maybe it will be better immediate scaling say chip at 2.8 all the way to a default 4.5 being their top part, and of course that 2.8 could be low clock bined similar power and thermal envelope to make that top end chip "sparkle"

Suppose there is just a ton of what ifs, but it now makes sense that all the AMD partners are working so close in tandem compared to the "Old Guard" and could vastly explain why Jim Keller seem to have went there instantly, maybe it was behind closed doors meetings and getting to the machine to start running, but now that all the bits and pieces are less likely out of whack, I am sure the differences in performance, value, and power use (and temperatures) will make 2016 an awesome year, I suppose with all the tech makers pushing so hard on multiple-fronts, and governments setting things into motion, its about damn time :p

This could be why Jim Keller "left" probably he be on contingency for redesigns etc, as this will be his personal moment to shine when it all comes together, especially if he had a major hand in all the places he has worked, to kind of "sync it all together"

Merry Christmas and Safe and Happy Holidays for all that walk through these digital doors at [H]ard Ocp staff, and guest alike :)
 
Last edited:

{NG}Fidel

Supreme [H]ardness
Joined
Jan 17, 2005
Messages
6,286
We can only hope that Jim left them with some winning designs and that this Samsung partnership gives them some type of advantage they can leverage. Both for GPUs and CPUs.
 

sirmonkey1985

[H]ard|DCer of the Month - July 2010
Joined
Sep 13, 2008
Messages
21,843
I believe GF & Samsung use the same manufacturing process, so it makes sense if AMD is to use them not use TSMC which would require a complete redesign.

There is not enough volume to warrant doing what Apple did IMO.
pretty much, GF decided to use samsungs manufacturing process given that it's already been proven and GF was way behind everyone else creating their own. there was an article about it i think 3 or 4 months ago, i'll see if i can find it and update my post.

nvm dragon basically beat me to it, should of read the rest of the thread..


as far as TSMC goes, i'm sure AMD's getting sick of being screwed by them anyways.. good to finally see a new manufacture(s) entering the gpu market.
 

Liger88

2[H]4U
Joined
Feb 14, 2012
Messages
2,657
I'm going to guess they'll use TSMC's more tried and proven 16nmFF+ initially on the more high-end cards and test 14nmLPP on the lower tier products until it matures. That extra bit of die-space from 14nmLPP could give AMD an advantage on the refresh.
 
Top