NVIDIA Kepler speculation thread


Nothing you can find on the internet at the moment. But I bet that if anyone else repeats it once it'll become just as valid as the XDR2 ram for Tahiti. Considering the interconflicting nonsense from various sites considering a hypothetical GTX680 (some thinking it's the top dog, some the performance SKU) has caused quite some confusion. Some go even as far thinking because it has a 2GB framebuffer that it'll most likely have a 512bit bus. The 512bit assumption isn't wrong per se with such a framebuffer size, it's just that 2GB ram doesn't make sense for a high end SKU these days. The 512MB difference between Cayman/2GB and GF110/1.5GB was/is bearable; a 1GB difference to the TahitiXT isn't.
 
Latest rumour is there will be no GK104 at all.. The named GK104 is nothing more than GK106.. 3 chips are GK110, GK106 and GK107, launch in April..

neliz was hinting the same thing earlier. The truth is we know nothing about Kepler. Chip names are appearing and disappearing at whim :D
 
He's wrong.

Wrong on what? April launch date or no GK104?
I have no doubt we won't see any Kepler before April.. GK106 could be exact half of GK110 like Cypress-Juniper, hence the 256bit/2GB rumours as fast as GTX580 or slightly faster.. GK106 prabably slower than GTX580 @same clocks but its much smaller and with 28nm it can be clocked higher and catch or pass it..
 
GK106 is not a performance part and there is of course a GK104. To steal from one source is plagiarism; to steal from many research.
 
Nothing you can find on the internet at the moment. But I bet that if anyone else repeats it once it'll become just as valid as the XDR2 ram for Tahiti. Considering the interconflicting nonsense from various sites considering a hypothetical GTX680 (some thinking it's the top dog, some the performance SKU) has caused quite some confusion. Some go even as far thinking because it has a 2GB framebuffer that it'll most likely have a 512bit bus. The 512bit assumption isn't wrong per se with such a framebuffer size, it's just that 2GB ram doesn't make sense for a high end SKU these days. The 512MB difference between Cayman/2GB and GF110/1.5GB was/is bearable; a 1GB difference to the TahitiXT isn't.

Well, they can always do 2GB and (presumably a bit later) 4GB SKU's, just as there are 1.5 and 3GB 580's. That's no biggie. 2GB is OK in my book, then again I dont mess with any of this multi-monitor stuff.

4GB on a single video card is kind of crazy to think about.
 
Just because some AIBs might have gotten some final schedules for the retail launch of GK110, GK106 and GK107, does not mean that GK104 does not exist, it might just be on another schedule...
 
Last edited by a moderator:
I doubt Nvidia would hide a codename from all but one partner.

It doesn't. I haven't seen any of the vendor roadmaps but I have the suspicion that some might state something like "GTX680/2GB/performance/high end". Absence of anything higher end than that, it should be easy to get confused and think that it's the top dog while it isn't. GK104 was and is planned to launch first and it's the highest end chip on any of those lists. GK110 comes later.

What is GK110 supposed to be - the high end part?

Yes.

Well, they can always do 2GB and (presumably a bit later) 4GB SKU's, just as there are 1.5 and 3GB 580's. That's no biggie. 2GB is OK in my book, then again I dont mess with any of this multi-monitor stuff.

Aren't the 3GB GTX580 vendor initiatives or am I wrong? Well you hit the nail on its head; IF Kepler should go for single SKU multi-monitor alternative you wouldn't want a high end GPU to run short of memory. Besides at those heights it's fairly impossible that the MSRP is below $500. Either equal or higher than that and I doubt 4GB would be too much.

4GB on a single video card is kind of crazy to think about.

While 2/3GB on upcoming performance GPUs won't be?
 
Isn't it likely that nVidia is just going to ship cards to AIBs as late as possible for them to slap on their stickers and drop them in boxes? This looks like another Fermi launch where nVidia had the cards pre-assembled and AIB's all received reference designs. That's one way to control the leaks and allow the FUD to spread unhindered.
 
Isn't it likely that nVidia is just going to ship cards to AIBs as late as possible for them to slap on their stickers and drop them in boxes? This looks like another Fermi launch where nVidia had the cards pre-assembled and AIB's all received reference designs. That's one way to control the leaks and allow the FUD to spread unhindered.

That still doesn't mean that a launch is imminent. Otherwise they would had revealed something at CES. According to fudzilla they intended initially to show at least something but decided not to and it was a pure business decision.
 
That still doesn't mean that a launch is imminent. Otherwise they would had revealed something at CES. According to fudzilla they intended initially to show at least something but decided not to and it was a pure business decision.

There's always the theory that whatever was going to be shown at CES wasn't showing much improvement over 7970 and hence they are holding off until drivers mature or they can bin more chips at higher clocks.

Regards,
SB
 
There's always the theory that whatever was going to be shown at CES wasn't showing much improvement over 7970 and hence they are holding off until drivers mature or they can bin more chips at higher clocks.

Regards,
SB

GTX680 has a GK104 afaik and that's a performance and not a high end chip. As to why they've decided to name it "680" Lord knows what marketing had in mind.
 
Back
Top