-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)
- Thread starter Boomstick777
- Start date
- Status
If this assumption by Guru 3D is correct it will still be many months before any cards are actually made.
It takes quite a while for these complicated chips to be created ( I cannot remember the exact timeframe but its one of those things that DM has explained before very well, maybe 'Hint Hint' he could do so again) and as we know very little in this industry is kept secret for very long. If GF were going to spring some super fast cards out for AMD I'm sure there would be more than just the forum babble to indicate that this was happening.
It takes quite a while for these complicated chips to be created ( I cannot remember the exact timeframe but its one of those things that DM has explained before very well, maybe 'Hint Hint' he could do so again) and as we know very little in this industry is kept secret for very long. If GF were going to spring some super fast cards out for AMD I'm sure there would be more than just the forum babble to indicate that this was happening.
Yes but I was meaning that it will be many months before Global Foundries produce any of the chips that this news (assumption) is talking about.
I hope so, but I haven't seen any Tape out news and that sort of thing is definitely not normally missed.
Yeah this is coming soon, Jan / Feb 2015, followed by cards based on new architecture in summer 2015.
Looking forward to new AMD cards, let's hope they are a big leap and not another round of incremental milking from both camps.
Unless redesigned a lot incremental is what happens on the same node (Intel aside)
efficiency goes up as well power draw goes down but we are about to hit a wall with die shrinks so they cant rely on that to do better from any brand.
14nm is when we are likely to spend 4 years on unless innovation happens.
I loaded popcorn and whiskey as once those who bought Nvidia cards get the buyers remorse
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Rules changed, significantly. I'm under the impression that tape outs at 20nm and below are significantly increased in time frame. IE 6 months was maybe normal at 40nm, maybe a tad longer at 28nm, it now might be closer to a year. More complex chips, having to move to double patterning which is where most of the issues with cost increase have come from.
http://www.extremetech.com/computin...hography-technique-to-push-moores-law-to-20nm
pretty good basic idea on it. current lasers are producing 193nm lightsources which do a surprisingly good job of etching 28nm chips, 20nm means double patterning and longer tape outs, more complexity in taping out, more yield problems, take longer to produce(ie chips might need 8-10 weeks rather than 6 weeks which effectively reduces throughput of wafers by a huge amount.
Less wafers with lower yields and higher costs... EUV will hopefully bring with it higher yields, less complexity, cheaper and higher volume(from the same fab size basically), which will all combine to make for significantly better value chip production. 20(and fake 14-16nm).
In terms of when these chips would be made, you need to know when they taped out and how long the tape out is.
GloFo have always been quiet in a general way. Charlie has had sources but aside from that few/any other people manage to call GloFo tape outs. TSMC is a leakfest which is the reason we find out about tape outs. The next gen of AMD gpu's could have taped out 12 months ago at GloFo and be about ready to launch, who knows.
I did see a nice design to product launch comparison from one node to the next including the estimates for 20-14nm... but I can't find them again. I think we were looking at 9-12months for most likely a TSMC tape out for 20nm and I think another month or two for 14nm finfets.
Anyway, there is no conclusive info which process or where they'd be made. There are really too many possibilities. 14nm finfet bulk Samsung/Glofo, 16nm finfet bulk TSMC, 20nm bulk TSMC, 28nm bulk GloFo or TSMC, or the newer 28nm FD-SOI GloFo. 28nm FD-SOI is going to be EXTREMELY competitive if not outright beat 20nm bulk and may not be far off bulk finfet processes. 20nm FD-SOI sounds like it's a pretty long time away and will likely not be available till 2016 and beyond.
Nvidia will almost certainly be making chips only at TSMC and only on 20 or 16nm, with AMD... it could be almost anything to be honest.
Reading between the lines from AMD's comments, it looks as if the upcoming interim GPU to take on the 970 / 980, will be based on 28nm but using HBM. So a combo of cutting edge tech and established node so that it's a cost effective solution until a new node transition is available later in 2015. Cooled by an AIO to keep it running nice and cool / quiet.
My guess is, upcoming card:
28nm.
HBM.
AIO Cooling.
3000+ Shaders.
Summer 2015 Cards:
20nm/14nm.
HBM.
Air cooling. (Except Dual card)
4000+ Shaders.
My guess is, upcoming card:
28nm.
HBM.
AIO Cooling.
3000+ Shaders.
Summer 2015 Cards:
20nm/14nm.
HBM.
Air cooling. (Except Dual card)
4000+ Shaders.
Soldato
I want the Dual card so i can go back to quadfire.
20nm and 14nm are not suitable for GPUs, we've been through this before...
The only option for a Summer launch is 28nm. There are now ways of making 28nm chips that are better than regular 20nm however. Anything after about July or so you should expect 16FF+ from TSMC. But if AMD are staying loyal to GF that leaves them without an option for a proper shrink.
The only option for a Summer launch is 28nm. There are now ways of making 28nm chips that are better than regular 20nm however. Anything after about July or so you should expect 16FF+ from TSMC. But if AMD are staying loyal to GF that leaves them without an option for a proper shrink.
Ha, so you think AMD won't have a die shrink at all. You crazy mate
. We'll see 20nm or 14nm FinFET (14LPP) used by AMD if they don't go with TSMC.Of course they will get a die shrink lol. So yeah,
My prediction is,
Upcoming card:
28nm.
HBM.
AIO Cooling.
3000+ Shaders.
Summer 2015 Cards:
20nm/14nm.
HBM.
Air cooling. (Except Dual card)
4000+ Shaders.
What's yours?
Hehe that counts out 99.93% of what's said on this forum then. (yes including an awful lot of what I say as well)
Literally every tech forum I read has people who work/ed in the semiconductor industry reporting you can't make a GPU on any 20/14nm process. The dies are 3x over the limit of what is currently acceptable size wise to be profitable. Too leaky. Yields are awful even for phone chips. Too much compromise to hit yield targets.
Meanwhile Boom is on here every day making "predictions" every time a new piece of clickbait pops up on wccf et al, based on nothing.
Meanwhile Boom is on here every day making "predictions" every time a new piece of clickbait pops up on wccf et al, based on nothing.
^^ So serious
Also I only made 1 prediction Mr serious, your just to scared to do the same
Boring..
Also I only made 1 prediction Mr serious, your just to scared to do the same
Boring..
- Status