Dunno really if it's 10 years already, but thanks for reminding me again that I've gotten that old....
Lol nah it's probably only been 4-5 years actually, but still if 2015 is the date that's a pretty long delayed chip - and most people think TSMC having 14nm by 2015 isn't very realistic either so lets say 2016 best case.
To answer your question though (why Logan would skip Denver) I suppose they don't want to change CPU and GPU at the same time? That makes some kind of sense as two large changes like that could set them back to square one again.
To answer your question though (why Logan would skip Denver) I suppose they don't want to change CPU and GPU at the same time? That makes some kind of sense as two large changes like that could set them back to square one again.
Gpu are faster for do this type of computing of an ARM or x86 based cores, so i dont see why deport this from the gpu shaders to an additional cores.
Now can it work on parrallel for speed up some calls.
A1xLLcqAgt0qc2RyMz0y
Veteran
January 2011 was not 10 years ago.
I should be more clear, my apologies. Nvidia might not have revealed it until then but it was well known about long before then.
2007 (didn't check for earlier) - http://www.eetimes.com/electronics-blogs/other/4410332/Nvidia-s-road-to-Denver-still-in-the-shadows
This has some pretty scary parallels to Bulldozer btw. Long rumoured, didn't quite make it for some reason, big reveal a couple of years later...long wait... now all that's left is for Nvidia to over-promise and under-deliver and it'll be Bulldozer all over again.
I think Denver was originally to be intro'd with 20nm, but I do believe (even though I'm not going to dig this up to verify it) that JHH said Tegra 6 (which will be the first Denver-based Tegra chip) will utilize fin-fet transistors, which means it has to be more advanced than 20nm.
Yep, just googled it. Denver will have finfet transistors, so it'll be 16nm.
Blazkowicz
Legend
So, what is the Tesla then?
the possibilities are :
- a Maxwell "20nm" with no included CPU cores
- a Maxwell "Finfet" with CPU cores
- a Maxwell "Finfet" with no CPU cores
And perhaps only the first one gets launched, or the first one followed by one of the two Finfet ones.
It's now apparent that when nvidia said "from cell phones to supercomputers" or something like that, they didn't litterally mean that everything had Denver or everything was a Tegra. So, it now feels probable that only Tegra 6 has Denver ; a Tesla with ARM solution may simply consist of selling racks which contain a Tegra 6 APU and Maxwell GPU (or m and n amounts of them), similar to the Tegra 3 + Fermi and Tegra 3 + Kepler devkits.
the possibilities are :
- a Maxwell "20nm" with no included CPU cores
- a Maxwell "Finfet" with CPU cores
- a Maxwell "Finfet" with no CPU cores
And perhaps only the first one gets launched, or the first one followed by one of the two Finfet ones.
It's now apparent that when nvidia said "from cell phones to supercomputers" or something like that, they didn't litterally mean that everything had Denver or everything was a Tegra. So, it now feels probable that only Tegra 6 has Denver ; a Tesla with ARM solution may simply consist of selling racks which contain a Tegra 6 APU and Maxwell GPU (or m and n amounts of them), similar to the Tegra 3 + Fermi and Tegra 3 + Kepler devkits.
Maxwell can't wait for finFET, since that's 2 nodes away (TSMC's 14nm process).
But nothing stops NVidia from making a 20nm Maxwell with Denver ARM core(s) using 20nm planar transistors. We know Parker will use 14nm finFET, but Parker is a mobile platform using Maxwell, not the Maxwell GPU architecture itself.
Blazkowicz
Legend
Needless to say I think we all expect a Maxwell generation of products on 20nm, at least the regular Geforces.
It has now come to my mind that they can sell a Maxwell Tesla variant based on a consumer chip, as they did with Tesla K10 (for folks who only care about FP32 performance without strong data integrity). So I'm curious whether the next "big" Tesla, successor to the GK110, is on 20nm or 14/16nm (that process is hard to name exactly, it's an ambiguous and hybrid process. Perhaps calling it the "FinFET" process is less ambiguous)
If we're interested into gaming cards : I don't think we need to care much at this point in time
we may look forward to 128bit, 192bit and 256bit consumer GPUs on 20nm to replace the current ones, giving more power efficiency and I don't know which other goodies.A January article I've found about the new process : it's hinting at a 20nm/14nm hybrid. Dunno if "16nm" is an artificial middle-ground to just give it a recognizable name..
And maybe there will be a 28nm/14nm hybrid?
http://lp-hp.com/blog/2013/01/17/first-silicon-at-14nm/
Needless to say I think we all expect a Maxwell generation of products on 20nm, at least the regular Geforces.
It has now come to my mind that they can sell a Maxwell Tesla variant based on a consumer chip, as they did with Tesla K10 (for folks who only care about FP32 performance without strong data integrity). So I'm curious whether the next "big" Tesla, successor to the GK110, is on 20nm or 14/16nm (that process is hard to name exactly, it's an ambiguous and hybrid process. Perhaps calling it the "FinFET" process is less ambiguous)
If we're interested into gaming cards : I don't think we need to care much at this point in time
A January article I've found about the new process : it's hinting at a 20nm/14nm hybrid. Dunno if "16nm" is an artificial middle-ground to just give it a recognizable name..
And maybe there will be a 28nm/14nm hybrid?
http://lp-hp.com/blog/2013/01/17/first-silicon-at-14nm/
Combine the last paragraph (which is an interesting twist if true for 28nm hybrids) with the rumors that some early =/<performance Maxwell variants may still first appear on 28nm.
LiXiangyang
Newcomer
Kepler K10 is pretty much a failure, Not only for its non-existing fp64 performance, the integer performance of K10 is also a diaster, I would say, its much worse than Fermi, so even if you dont care fp64, K10 is still not a wise investment, and it still require the level of ILP of K20 to achieve peak performance.
So basically you get a card, which is cost like a K20, and need a K20 level effects to achieve sub-Fermi level performance.
So basically you get a card, which is cost like a K20, and need a K20 level effects to achieve sub-Fermi level performance.
The way they are talking they make it sound like they solved 20 nm ... when no one but Intel has produced any real chip yet. Don't expect 20 nm this year, don't expect anything better for 3 ... from anyone except for Intel of course.
Blazkowicz
Legend
It's been quite a while since they've been saying "14nm comes very shortly after 20nm".
So I expect 20nm in 2014 and 20/14 in 2015.
So I expect 20nm in 2014 and 20/14 in 2015.
i think this is the right place (?)..however: http://semiaccurate.com/2013/05/20/nvidias-volta-gpu-raises-serious-red-flags-for-the-company/
