-
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.
The next gen xxnm GPU thread.
- Thread starter Kaapstad
- Start date
TSMC plans to use EUV lithography for 5nm.
10nm is on-track for volume production start in Q4 2016.
We have taped out our first FinFET products.
Given how much difficulty they've had shrinking beyond 28nm, I'm not at all sure why people aren't expecting massive delays for future process shrinks. 10nm in 2017? I'll be surprised if we see it before 2020.
Unless they are further distorting reality with the nomenclature, and "10nm" is really 22nm or something.
Unless they are further distorting reality with the nomenclature, and "10nm" is really 22nm or something.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,488
You only have to look back at when tsmc announce stuff and when AMD/nvidia get products out... Tsmc have already announced 16nm is in volume, but 16nm GPU's are still 9-12 months away, if tsmc say 10nm is q4-2016 to q1-2017, then its going to be 2018 before we see GPU's, so yeah we are looking at 2017 16nm refreshes
I think we will see quite moderate mm2 or density die for Nvidia, even with 'big' pascal. They have no reason to push the limits too hard (although the margins and their appeal grants them additional leeway), what they need is a high perf dp part to replace their aged GK110/210 reasonably soonish.
So 2017 16nm refresh could have plenty of untapped performance to be realised.
So 2017 16nm refresh could have plenty of untapped performance to be realised.
Pascal GPU has 17 billion transistors.
Last edited:
What they are really slow with is Little Pascal which will come first and pack around 9 to 10 billion transistors.
What they are really slow with is Little Pascal which will come first and pack around 9 to 10 billion transistors.
8 billion + 1
Not really related to GPUs but still possibly worth a read.
SS adds 10nm FinFET tech to roadmap, mass production in 2016.
SS adds 10nm FinFET tech to roadmap, mass production in 2016.
Why do GF suck so bad at making their own processes... 28nm late, 20nm MIA, 14nm saved by Samsung - have to wonder if AMD will throw their hands up and take the hit from the supply agreement penalty or just dump them @ 10nm in favour of SS/TSMC.
I read a theory last night which I have an eerie feeling will come true, AMD competing against 16FF+ Pascal with tiny but dense/efficient 14nm chips because GF/SS can't offer anything that can do 300+mm2. That would be short sighted as NV will just beat them with raw grunt. I had assumed that a high-performance process would be offered by the time AMD taped out but there's no sign of one, so unless it's a secret I guess this theory might just pan out.
I read a theory last night which I have an eerie feeling will come true, AMD competing against 16FF+ Pascal with tiny but dense/efficient 14nm chips because GF/SS can't offer anything that can do 300+mm2. That would be short sighted as NV will just beat them with raw grunt. I had assumed that a high-performance process would be offered by the time AMD taped out but there's no sign of one, so unless it's a secret I guess this theory might just pan out.
Little Pascal first would be what we would expect, especially with what NVidia have done for the last few generations. But I've got a funny feeling that they will surprise us this tome round, what with the rumour that big Pascal taped out first, we might see a bonkers powerful Pascal Titan with an astronomical price first along with the compute variant, then the little Pascal before mainstream big Pascal variant.