• 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.

Kepler high-end (possibly) delayed until Q3/4 2012

kissenger

kissenger

kissenger

Associate
Joined
10 Apr 2008
Posts
2,491
According to this article...

http://www.techpowerup.com/155727/NVIDIA-GeForce-Kepler-Roadmap-Compiled.html

133a.jpg


Doesn't seem overly-credible at the moment, so take it with a pinch of salt.
 
WHy not overly credible, nothing has stated its coming before mid year. GK107, consider the specs listed and the name(for Fermi high end was GF100 and super ultra crap low end was GF108, GK107 is low end).

GK107 is the only thing confirmed to have taped out, the big tape outs almost always leak, tape out to product launch and availability is usually 4-6 months. GK110 not taped out yet, not going to be out before April realistically, and likely much later than that already.

Why they are calling it GK104, don't know, its probably a change to fit in the GF110 dual card, and with the GK107 and the disappearing of the low end in general and capacity issues...... product line ups are changing for AMD and Nvidia.

The only story really talking about high end Kepler to date is suggesting mid 2012 already, this merely adds to the credibility of that previous story.

Its a bit odd, it does somewhat suggest a line up that COULD be along the lines of in terms of fermi, only having the 560ti out at first, then releasing a dual version then waiting till 2013 to release the "full" 580gtx. I think GK104 is going to be noticeably bigger than a 560ti but less than a 580gtx equivilent, with that only due later on.

It also suggests the possibility the "real" GK102 was a bit 480gtx's and came back non working and got cancelled rather than go through multiple respins at great expense before a full base layer respin to take it to the 580gtx.
 
Last edited:
If that's true (which I have my doubts about) AMD has a good opportunity to gets its high end 28nm out the door and nab more of the discrete video card market share away from Nvidia. Realistically though if Nvidia can't get their high card card out until Q3 2012 you would have to think AMD will have silmilar problems with the 28nm process.

Other stories have been floating about Nvidia releasing some sort of 28nm product next month (even less likely give Chinese New Year and the logistics involved) so we won't have to wait long to see if this story is true or not. The only thing we know for sure is Nvidia did expect to be launching Kepler in 2011 which is confirmed from press a press conference held by their CEO last year.
 
Have to wonder where it all went wrong... even if that is fake theres definitely some delays in the pipeline... GK107 was _supposed_ to have taped out 4-5 months ago.


Freddie1980 said:
Other stories have been floating about Nvidia releasing some sort of 28nm product next month (even less likely give Chinese New Year and the logistics involved) so we won't have to wait long to see if this story is true or not.
Click to expand...

Can't see that being likely even tho nVidia has been claiming a 28nm part would be released around November for quite some time - as far as I'm aware the GK107 tape out is relatively recent - which would put us still atleast 2-3 months absolute minimum away from a new product on the market.
 
Last edited:
It looks like
580GTX replacement is the GK104
590GTX replacement is the GK110 (dual GK104)
GK112 very late 2012, early 2013 is a 585GTX?
 
I was wondering when that would be posted here.

Not sure on the credibility of the source, but I do think that it will be early summer before the decent performance stuff arrives.
 
But that anything above a 550ti will require PCI 3.0?0 Hardware manufacturers don't like rapid change on platforms (look at USB 2>3) there is still a VERY large market out there of PCI 2.0 boards with unsaturated lanes, so a true need to go 3.0 isn't there.

I could foresee it in some down the line revisions but not at launch.

Unless 3.0 is backwards compatible to 2.0 (I dunno havn't done much reading) in which case this entire post is somewhat foot-in-mouth.
 
Rroff said:
Have to wonder where it all went wrong... even if that is fake theres definitely some delays in the pipeline... GK107 was _supposed_ to have taped out 4-5 months ago.




Can't see that being likely even tho nVidia has been claiming a 28nm part would be released around November for quite some time - as far as I'm aware the GK107 tape out is relatively recent - which would put us still atleast 2-3 months absolute minimum away from a new product on the market.
Click to expand...

I think GK107 only taped out a couple months ago, the "earlier" tapeout which guys like Fud suggested were Kepler by talking about 28nm tape outs and kepler in the same article, I think turned out to be either 28nm shrinks of some mobile bits and bobs... or they are rebrands and Fud just got confused about everything, which isn't uncommon :p

As for Gk107, it is Nvidia so a respin isn't the most unlikely of situations in all honesty. While their new low end normally doesn't require a respin, thats normally because they've done the high and midend by that point so have a LOT more experience, this time the GK107 seems to be the "tester" part.

Its all a tad odd, because Gk104 is mentioned at being around the 250W mark........ now either that's real, or that's an Nvidia 250W, so 480gtx and more like 300W. That really doesn't suggest a small part to me, in fact from a supposedly woeful 40nm with not a whole lot added to it to fight leakage, to their first HKMG process IIRC(or did 40nm have that) which should tackle leakage well, you'd have thought coming in with a 70-80W performance part might even drop power usage.

But the GK104 seems set up to be, not exactly a truly high end part....... mostly basing it off 384 bit bus vs GK112's 512bit bus......... if the bus scales with core size it could hit at something a decent wedge smaller..... but then how is it using 250W?

also yes, pci-e 3 is backwards compatible.
 
Looks like GK110 is two GK104 chips. Possibly two chips on a single die or some such scheme, or perhaps, more likely (since they're calling it GK110), they are fabricating two GK104 GPUs on the same chip. And then the one above it -- GK112 looks like a single monolithic GPU. It has a shorter (but singular) memory bus, and it is still placed above the other. Perhaps it has more CUDA cores?
This is unprecedented for commercial GPUs afaik. Perhaps they find their GK104s are just so good at scaling in a multi-processor config that it is worthwhile to put two of them rather than design a bigger processor as teh stopgap between this and their GK112.

Or perhaps something is amiss in the info provided.




edit: Right, just read the article. So they are going with SLI to compete on the higher end. Sounds like they are fighting AMD on their front now.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom