There's kindof news:
http://wccftech.com/nvidia-maxwell-...eforce-800-series-arrives-2014/#ixzz2tKy20yWr
It's not confirmed that the Nvidia 800 series is 20nm, but there's pretty much no other reason they would wait until 2H 2014. The timing would be a pretty major coincidence given the state of TSMC's 20nm process.
No word at all I've seen on 20nm AMD.
Thanks, looking at it now. I have looked at TSMC's roadmap and I believe they are on mass production so I am having a long shot and guessing that their should be plenty of wafers to go around soon.
You need to look into it further. They have different levels of 20nm. GPU's are among the most complex chips manufactured, and are always the last to finally be produced.
The vast majority of 20nm chips are going to phone manufacturers. I am standing by my 4Q14 prediction for 20nm GPU's. Although I could see a 3Q release happening if things go well.
8-10 months. The 20nm they have starting now is not high performance.
What 20nm?There is no 20nm Graphic cards right now.
What 20nm?There is no 20nm Graphic cards right now.
Thought I would give this a bump to see what has happened recently.
Bad news probably.
http://forums.anandtech.com/showthread.php?t=2377192
Interesting, this slide implies that the A7's die costs 50$, which seems quite a lot. I wonder how everyone's going to improve performance if transistors will cost the same for the next 4 years or so. A 20nm GTX 780Ti equivalent could have 2 times as much transistors because of the higher density, but it would cost 2 times as much (>600$ for the die alone).TSMC's 20 nm isn't increasing density as much as the name suggests add to that the costs increase for the new process, and you get a wavy curve.
The chips are actually just as expensive to produce, so there is very little incentive to switch and cards are ridiculously expensive as it is. The slide is from ARM, but you see how the number of transistors a dollar can buy peeks at 28nm. (btw. 20 million a dollar, that's ~350 $ for a 780 ti GPU)
If transistors aren't getting cheaper in 2015 then neither are GPUs or GPU performance/$, unless we see some rabbit-out-of-the hat architecture improvements.
![]()
So you mean that ARM is wrong by a factor of more than 2.5?! Source?