AMD and Nvidia 20nm and 16nm GPUs Slightly Delayed

fanboy

[H]ard|Gawd
Joined
Jul 4, 2009
Messages
1,057
This has notting to do with what is coming from AMD and what is the 380x as it is built on a GF 28nm and will offer 20 to 25% across the board in power savings and performance just switching 28nm fabs without adding anything Tonga could add to performance.. Hawaii 2.0 could end up being 30 to 40%faster then the current 290x.

AMD’s 28nm GPUs are based on GlobalFoundries’ 28 SHP Node – Alleges Italian Report

Read more: http://wccftech.com/amd-gpus-28nm-shp-glofo/#ixzz3NLRMFvgt
 

OmgitsSexyChase

Limp Gawd
Joined
May 1, 2013
Messages
194
This has notting to do with what is coming from AMD and what is the 380x as it is built on a GF 28nm and will offer 20 to 25% across the board in power savings and performance just switching 28nm fabs without adding anything Tonga could add to performance.. Hawaii 2.0 could end up being 30 to 40%faster then the current 290x.

AMD’s 28nm GPUs are based on GlobalFoundries’ 28 SHP Node – Alleges Italian Report

Read more: http://wccftech.com/amd-gpus-28nm-shp-glofo/#ixzz3NLRMFvgt

Is the 390x Going to be 20? if not GG, Im so sick of waiting Nvidia and AMD have been projected 20nm for ages just to be delayed and delayed.....I don't even care about 16..let me see 20
 

chenw

2[H]4U
Joined
Oct 26, 2014
Messages
3,977
What exactly is in the 20nm that makes people want it more and more?
 

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
20,183
Smaller process means more transistors can be packed into the same size silicon die and power consumption per transistor reduces.
Shorter distances and smaller transistors permit higher frequencies to be used.
There will be other improvements and new issues to resolve.

For instance, electron leakage increases making the silicon use more power and get hotter than it otherwise would.
This can be bad enough to negate the efficiency gains and reduce the projected maximum frequency.
There is more heat generated within the same size die so cooling can become a problem.
(note that despite transistors being smaller, the transistor density is higher. There are more of them resulting in net higher power use per unit area)
 

chenw

2[H]4U
Joined
Oct 26, 2014
Messages
3,977
Right, so basically it's just performance increase.

I thought I had missed something magical about the 20nm process.
 

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
20,183
Erm, that is the holy grail as well as power use reduction for the same amount of processing.
Smaller size dies can be useful too.
What else did you imagine could be improved?
 

LordEC911

[H]ard|Gawd
Joined
Jul 7, 2013
Messages
1,461
Except that the 20SOC process at TSMC and the two known 20nm processes at GF are both LP typically used for mobile and networking ASICs. Not something you would expect to see a GPU built on.

As for TSMC's 20nm planar process, it is pretty dang expensive for double the density but only a 15-20% power/speed benefit. Customers are looking at paying ~2x in upfront fixed costs and at least another ~30-40% more for wafers. Toss in the normal yield issues on a new node and you come to the conclusion that unless you are a major player, Qualcomm and Apple, that need every little benefit and buy huge numbers of wafers you are much better off waiting for FinFet.
 
Last edited:

Factum

2[H]4U
Joined
Dec 24, 2014
Messages
2,466
It is sad how PR have gotten into "process"-land.

After Intel went 22nm and on track to 14nm all others seem to scramble for the PR-nm train.

Sudden all sorts of excotic process appeared. "20nm" "16nm" "16nm+".

Not all process are born equal and Intel's lead seems to be increasing, if you can see through all the false marketing/PR from GF, TSMC
etc.
 

TaintedSquirrel

[H]F Junkie
Joined
Aug 5, 2013
Messages
11,656
So we're looking at possible Titan 2, 980 Ti, 960... and a new series of at least 2 GPU's from AMD Bermuda and Fiji...in Q1/Q2, all on 28nm.

Then we have nothing for 1 year until 20nm at the "end of 2015" at the earliest, assuming there aren't more delays. Snoooooze.
780 Ti to 980 was a little under a year so I guess we're seeing a repeat of that.
 

LordEC911

[H]ard|Gawd
Joined
Jul 7, 2013
Messages
1,461
It is sad how PR have gotten into "process"-land.

After Intel went 22nm and on track to 14nm all others seem to scramble for the PR-nm train.

Sudden all sorts of excotic process appeared. "20nm" "16nm" "16nm+".

Not all process are born equal and Intel's lead seems to be increasing, if you can see through all the false marketing/PR from GF, TSMC
etc.

Lawl. Intel was the one to start it...

Anyway, Intel's lead has shrunk as the field thins out.
EUV will pretty much even the playing field for bulk, though Intel will likely still have a lead.
 

Liger88

2[H]4U
Joined
Feb 14, 2012
Messages
2,657
Lawl. Intel was the one to start it...

Anyway, Intel's lead has shrunk as the field thins out.
EUV will pretty much even the playing field for bulk, though Intel will likely still have a lead.


But at least Intel's methodology for shrinks is a little more....standard? I guess that would be the word. TSMC has completely bastardized it into meaning nothing after 28nm because of significant delays and FinFet issues (developing their own 3D structure). IIRC the 16nm FinFet+ is basically the equivalent to actual 16nm while the normal 16nm TSMC process is closer to 20nm planar and that's only as reference to performance not actually shrinkage.

Actually, I'm going to not start lol. It has gotten to a point that it leads to people now debating that not everything is at XXnm claimed inside the chip being built. It's gotten stupid and Nvidia has proved a new node shrink isn't always necessary in Maxwell. They can push 50% performance on the same node if they want over the previous design.
 

harmattan

Supreme [H]ardness
Joined
Feb 11, 2008
Messages
5,057
Right, so basically it's just performance increase.

I thought I had missed something magical about the 20nm process.

Not just performance. Historically, in addition to the really big leaps in performance, smaller processes have brought significantly lower power consumption and heat.
 

Uthur

n00b
Joined
Dec 18, 2014
Messages
2
So all the 20NM rumors are rong about R9 3xx? Are we still getting HBM? i wonder if some series will be 28nm like Tonga and Hawaii 2.0** and then the Fiji series could be 20nm or 16NM??? I mean if the 390 releases in Summer like suspected thats months after Tonga XTX and Hawaii 2.0 sothere is time right? I want to see a real increase from my 290 so i can give it to my Dad and get 3xx.
 

fanboy

[H]ard|Gawd
Joined
Jul 4, 2009
Messages
1,057
GF is going to impress with there 28nm which offer the fastest transistors you can buy .. the 380X is going to be fast..
 

lostinseganet

[H]ard|Gawd
Joined
Oct 8, 2008
Messages
1,197
Not just performance. Historically, in addition to the really big leaps in performance, smaller processes have brought significantly lower power consumption and heat.
This is true. The company has to screw up on the card to not achieve this.
 
Top