More 20nm news

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
Proof in the pudding. AMD will have no 20nm products at all in 2014. It sounds like there have been zero 20nm tape outs on top of it.

Christopher Rolland - FBR Capital Markets
Hey guys. Congrats on the quarter and thank you for letting me ask question as well. Can you guys talk about the puts and takes of either moving or perhaps not moving your next-generation follow-on to Kaveri to the next node here and how this might affect your decisions to move other products and families to the next node, or not to move them? Thanks.

Lisa Su - SVP and General Manager of Global Business Units
Sure, Chris. So let me take that and give you a little bit of our thinking.

So in terms of product and technology selection, certainly we need to be at the leading-edge of the technology roadmap. So what we've said in the past is certainly this year all of our products are in 28-nanometer across both, you know, graphics client and our semi-custom business. We are, you know, actively in the design phase for 20-nanometer and that will come to production. And then clearly we'll go to FinFET. So that would be the progression of it.

Relative to the competitiveness of the products, we feel very good about the competitiveness of the products, to vary in terms of total compute, what we're able to do across the stack with our Beema product line and our graphics capability.

http://seekingalpha.com/article/214...arnings-call-transcript?page=5&p=qanda&l=last
 
Feb 19, 2009
10,457
10
76
I think we're all expecting a 2015 (mid) 20nm launch of GPUs. 2014 is just too wildly optimistic given the past leaks, signs and now more news.
 

wand3r3r

Diamond Member
May 16, 2008
3,180
0
0
While it does appear 2015, and I'm inclined to believe it will be, AMD said 7970 will remain the highest card vs. the titan and later introduced the 290/x.

All signs point to 2015, but at some point they want to keep selling cards so I don't think they will let us know until as late as possible just to maintain sales.
 

n0x1ous

Platinum Member
Sep 9, 2010
2,574
252
126
Intel - please open your glorious fabs to nvidia and amd so we don't have to suffer through TSMC delays anymore.
 

Face2Face

Diamond Member
Jun 6, 2001
4,100
215
106
Do you guys think it's possible that Nvidia will release more Maxwell based cards on 28nm this year? Maybe some higher performing ones?
 

n0x1ous

Platinum Member
Sep 9, 2010
2,574
252
126
Im sure they could do GM104 on 28nm. bump the die size up 50mm2 and optimize like GM107 and have a pretty good outcome. Certainly they need 20nm for GM100
 

Techhog

Platinum Member
Sep 11, 2013
2,834
2
26
I don't see why they would given news like this?

Yeah. AMD can't compete.

It's looking more and more like we'll have to get used to waiting 3-4 years between generations of GPUs. This is going to make APIs like Mantle and DX12 even more important.
 

dangerman1337

Senior member
Sep 16, 2010
337
5
81
Do you guys think it's possible that Nvidia will release more Maxwell based cards on 28nm this year? Maybe some higher performing ones?
Since Nvidia have said they are going to release higher performing ones (Second Generation Maxwell) this year I'm betting there won't be any 20nm any GPUs from AMD and Nvidia and Pirate Islands and Second Generation Maxwell will be on 28nm and we'll see 16FF GPUs mid to late 2016 (Pascal and AMD equivalent) as that will be a cost effective jump by then. I think AMD partners also expect a fall release for Pirate Islands as well (someone said this at Cebit).

20nmSoC is too cost-ineffective to make high-end GPUs on and Maxwell shows big jumps on 28nm and hopefully Nvidia and AMD can do the same level of jump or better with second generation Maxwell with GM200, GM204 and GM206 like GK107>GM107. Also this post on Beyond3D has someone claiming that Nvidia will not be doing any 20nm GPUs at all: http://forum.beyond3d.com/showpost.php?p=1837191&postcount=1478, http://forum.beyond3d.com/showpost.php?p=1838346&postcount=1507

I still think we'll see a jump due to architectural improvements and Die shrinks don't account much for increases in performance*.

EDIT: * These days.
 
Last edited:

dangerman1337

Senior member
Sep 16, 2010
337
5
81
Wow, 3 and a half years between node changes.
It'll be more like 4 years, while sounding bad I think it will be for the better as getting straight on Die shrinks is not economical at all and we'll see a tighter release deadline in the future instead of 1+ year for a big one after a performance range masquerading as an Enthusiast one it'll be a few months. IMO for Nvidia I predict mid to late 2016 for Pascal on 16nmFF, 2H 2018 for Volta on 10nm and not sure after that though, maybe Einstein on 7nm in late 2020?
 

blackened23

Diamond Member
Jul 26, 2011
8,548
2
0
:eek: errrrr-yeah I'm pretty sure they do. Thats why this matters so much....

Let's define what "die shrink" means.

It's a correct statement. A "die shrink" in the old sense simply means taking an old architecture and applying it to a new node with ZERO CHANGES; this is no longer economically feasible and simply plopping an old architecture on a new node doesn't increase performance. ATI has done this in years past and their motivation to do so was financial - in the PAST, you received more chips per wafer and thus more profit. But this is no longer the case because of higher wafer costs. Therefore AMD/NV can't simply do die shrinks anymore. IIRC, ATI did a die shrink with the R600 to lower costs and increase profits. This was common place in years past. Now? It isn't because it isn't economically viable, especially with 20nm. And it doesn't change performance characteristics.

What DOES increase performance is utilizing the increased density with a new architecture with more transistors. This gives a meaningful performance increase. Therefore you will see both NV and AMD apply new architectures to the 20nm node, 2nd gen Maxwell for NV and whatever AMD has for AMD. It won't be a "die shrink". A die shrink would be NV putting Kepler on 20nm. Or AMD putting Tahiti / Hawaii on 20nm with zero changes. Back in 2006, this would have happened. Now, it can't because of economic feasibility.
 
Last edited:

n0x1ous

Platinum Member
Sep 9, 2010
2,574
252
126
Let's define what "die shrink" means.

It's a correct statement. A "die shrink" in the old sense simply means taking an old architecture and applying it to a new node with ZERO CHANGES; this is no longer economically feasible and simply plopping an old architecture on a new node doesn't increase performance. ATI has done this in years past and their motivation to do so was financial - in the PAST, you received more chips per wafer and thus more profit. But this is no longer the case because of higher wafer costs. Therefore AMD/NV can't simply do die shrinks anymore. IIRC, ATI did a die shrink with the R600 to lower costs and increase profits. This was common place in years past. Now? It isn't because it isn't economically viable, especially with 20nm. And it doesn't change performance characteristics.

What DOES increase performance is utilizing the increased density with a new architecture with more transistors. This gives a meaningful performance increase. Therefore you will see both NV and AMD apply new architectures to the 20nm node, 2nd gen Maxwell for NV and whatever AMD has for AMD. It won't be a "die shrink". A die shrink would be NV putting Kepler on 20nm. Or AMD putting Tahiti / Hawaii on 20nm with zero changes. Back in 2006, this would have happened. Now, it can't because of economic feasibility.

excellent clarification - yes thats pretty much what I mean when I say "die shrink" the space for more transistors on a new arch.
 

know of fence

Senior member
May 28, 2009
555
2
71
Wow, 3 and a half years between node changes.

Huh? When AMD rushed to market with the 7970s in early 2012 it was clear that we'll be stuck on 28nm for a while. It's been 2 years and a quarter since, early 2015 will be 3 years.

http://www.anandtech.com/show/5312/amd-radeon-hd-7970-now-for-sale

TSMC began volume production in 2014, the next Iphone will have a A8 20nm SoC, and those come out around September. Maybe late this year is still a possibility, the phrasing in this call is somewhat vague "So what we've said in the past".
 

dangerman1337

Senior member
Sep 16, 2010
337
5
81
:eek: errrrr-yeah I'm pretty sure they do. Thats why this matters so much....

Let's define what "die shrink" means.

It's a correct statement. A "die shrink" in the old sense simply means taking an old architecture and applying it to a new node with ZERO CHANGES; this is no longer economically feasible and simply plopping an old architecture on a new node doesn't increase performance. ATI has done this in years past and their motivation to do so was financial - in the PAST, you received more chips per wafer and thus more profit. But this is no longer the case because of higher wafer costs. Therefore AMD/NV can't simply do die shrinks anymore. IIRC, ATI did a die shrink with the R600 to lower costs and increase profits. This was common place in years past. Now? It isn't because it isn't economically viable, especially with 20nm. And it doesn't change performance characteristics.

What DOES increase performance is utilizing the increased density with a new architecture with more transistors. This gives a meaningful performance increase. Therefore you will see both NV and AMD apply new architectures to the 20nm node, 2nd gen Maxwell for NV and whatever AMD has for AMD. It won't be a "die shrink". A die shrink would be NV putting Kepler on 20nm. Or AMD putting Tahiti / Hawaii on 20nm with zero changes. Back in 2006, this would have happened. Now, it can't because of economic feasibility.
Thanks for saying what I was saying, they used to but nowadays they do not do much compared to architectural changes.

Huh? When AMD rushed to market with the 7970s in early 2012 it was clear that we'll be stuck on 28nm for a while. It's been 2 years and a quarter since, early 2015 will be 3 years.

http://www.anandtech.com/show/5312/amd-radeon-hd-7970-now-for-sale

TSMC began volume production in 2014, the next Iphone will have a A8 20nm SoC, and those come out around September. Maybe late this year is still a possibility, the phrasing in this call is somewhat vague "So what we've said in the past".
As I have explained GPUs on TSMC's 20nmSoC (there is no 20nm HP) is very unlikely, by the time it becomes cost-effective we'll have 16nmFF that will be similar in costs to 20nmSoC by TSMC which is mostly 20nm with FinFETs which will likely scale.

I expect something like a GP107 on 16nmFF (will be economical to create due to small size) on H1 2016 and Pascals that perform better than GP107 in 2H 2016.
 

VulgarDisplay

Diamond Member
Apr 3, 2009
6,188
2
76
I don't feel as stupid for upgrading to a r9 290 from my reference 7970. Looks like I'll be using it for quite some time.
 

f1sherman

Platinum Member
Apr 5, 2011
2,243
1
0
why is this a surprise?
remember CEO Rory saying AMD will be carefull about jumping on new nodes
and AMD has not even been mentioned among those who reserved TSMC 20nm capacity

skipping 20nm and jumping straight to 16nm has been in speculations for quite some time
this is just a confirmation
 

Qwertilot

Golden Member
Nov 28, 2013
1,604
257
126
2016 sounds a very logical time scale to me from a demand point of view. Monitors are doing very odd things just now of course - instead of a smooth progression from 1900* 1080 to 2500*1400 - they're seemingly jumping all the up out to 4k in one go.

That certainly doesn't incentive NV/AMD to rush and push out a mildly faster new range of cards. I'll certainly be waiting until something like a single 760 equivalent can push 4k reasonably, and I doubt I'll be alone.