NVIDIA Kepler speculation thread

So the Kayla board announced today has a GPU, which, according to AnandTech, is a 1 SMX Kepler chip. Does the GK208 have anything to do with that chip?
 
Also interesting:


D14 was Geforce 600.
GK110 relaunch as GTX 780/770?

It can keep the "GTX Titan" name while Geforce 700 series get launched as "lesser" products.
A Titan with one more SMX disabled and 320bit might end up with a 7xx name such as GTX 775 w/ 2.5GB, maybe.
 
It can keep the "GTX Titan" name while Geforce 700 series get launched as "lesser" products.
A Titan with one more SMX disabled and 320bit might end up with a 7xx name such as GTX 775 w/ 2.5GB, maybe.

If there will be 7xx-series from Kepler-based products, they'll be rebrands or possible clock-tweaks, new chips? Not going to happen (besides the GK208 which could be the Kayla-chip)
 
BqV9gCh.jpg


I wonder if the other Kepler Refresh GPUs also are Compute Capability 3.5(GK110) like GK208.
 
As Kaotik said they can do new brandings and not do any new chip.
Though, a new branding and clock is sometimes decent (9800GT or 9800GTX to GTS 250) or entirely pointless (8800GT to 9800GT).

The high end would look weak I think, with GK104. A GK104 GTX 780 looks pointless, or doesn't have very good value : they should do a GK204, which I speculate to have 10 SMX and a 384bit bus, and no ECC in the caches (etc.)

Geforce GTX 700 series would thus use GK208 (and possibly GK107), GK106, GK104 (even if some "ti" model), GK204.

I have of course no grounds to base this on. Rationale for not doing a GK204 is AMD keeps its Tahiti at high end, and nvidia thinks a very small clock boost on GK104 is enough.
 
As Kaotik said they can do new brandings and not do any new chip.
Though, a new branding and clock is sometimes decent (9800GT or 9800GTX to GTS 250) or entirely pointless (8800GT to 9800GT).

The high end would look weak I think, with GK104. A GK104 GTX 780 looks pointless, or doesn't have very good value : they should do a GK204, which I speculate to have 10 SMX and a 384bit bus, and no ECC in the caches (etc.)

While I think it would be a great chip if it were to happen, I just don't see Nvidia using their engineers to create a new Kepler high end gaming chip. There is another GK110 Geforce card coming this summer: http://translate.google.com/transla...nedbantat-geforce-gtx-titan-med-gk110&act=url.

A 10 SMX 384-bit chip would probably exceed this card's performance because of better bandwidth and higher core clocks / ROP throughput and also encroach on Titan territory to the point of making Titan look absolutely worthless (instead of just way overpriced). Nvidia has 4 Kepler chips, I don't see them creating a fifth one just to sell to gamers. I think the cutdown GK110 in the link I provided will be the gtx700 series flaghship (gtx780 or gtx785). GK114 / GK204 (whatever Nvidia calls GK104's refresh) will probably just be transistor optimization tweaks, nothing extraordinary. I expect about 5-7% faster than gtx680 with a slightly lower or same TDP with the gtx770 label.
 
Last edited by a moderator:
While I think it would be a great chip if it were to happen, I just don't see Nvidia using their engineers to create a new Kepler high end gaming chip. There is another GK110 Geforce card coming this summer: http://translate.google.com/translate?sl=auto&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&eotf=1&u=http%3A%2F%2Fwww.sweclockers.com%2Fnyhet%2F16711-nvidia-forbereder-nedbantat-geforce-gtx-titan-med-gk110&act=url.

A 10 SMX 384-bit chip would probably exceed this card's performance because of better bandwidth and higher core clocks / ROP throughput and also encroach on Titan territory to the point of making Titan look absolutely worthless (instead of just way overpriced). Nvidia has 4 Kepler chips, I don't see them creating a fifth one just to sell to gamers. I think the cutdown GK110 in the link I provided will be the gtx700 series flaghship (gtx780 or gtx785). GK114 / GK204 (whatever Nvidia calls GK104's refresh) will probably just be transistor optimization tweaks, nothing extraordinary. I expect about 5-7% faster than gtx680 with a slightly lower or same TDP with the gtx770 label.

But at what price point ? Untill they decide to decrese the price of Titan, + the second slimmed GK110, this will put a GTX770 at 550+ $..
nobody will buy a GTX770 with 5-7% more performance of a 680 ( so in the overclocked retail 680 territory) for 550$.

GTX770 = 550$, GTX 780/GK110 = 700$ and Titan 900$ ? Because i dont see them release a cut down version of Titan GTX780 @ 550-600$ for get their 770 at 450$ with the full titan still hovering at 900$ .

I mostly think they will release the slimmed version outside GTX 700 series.. for offer a lower priced Titan, but still keep the card at something around 750 $ .
 
Last edited by a moderator:
While I think it would be a great chip if it were to happen, I just don't see Nvidia using their engineers to create a new Kepler high end gaming chip. There is another GK110 Geforce card coming this summer: http://translate.google.com/transla...nedbantat-geforce-gtx-titan-med-gk110&act=url.

A 10 SMX 384-bit chip would probably exceed this card's performance because of better bandwidth and higher core clocks / ROP throughput and also encroach on Titan territory to the point of making Titan look absolutely worthless (instead of just way overpriced). Nvidia has 4 Kepler chips, I don't see them creating a fifth one just to sell to gamers. I think the cutdown GK110 in the link I provided will be the gtx700 series flaghship (gtx780 or gtx785). GK114 / GK204 (whatever Nvidia calls GK104's refresh) will probably just be transistor optimization tweaks, nothing extraordinary. I expect about 5-7% faster than gtx680 with a slightly lower or same TDP with the gtx770 label.

There probably won't be GK114 or GK204, but just Maxwell in 2014, just like AMD isn't making new "refresh line".
OEM rebrands might happen.
 
So, Geforce 7xx could be a "lost generation", like Geforce 3xx, and then Maxwell could be the Geforce 800 series.
I'd say that's certainly a possibility, especially if Maxwell initially launches on 28 nm early in 2014 (there seems to be little point for a "GK204" in late 2013 if "GM104" shows in Q1/Q2 2014). If GK208 comes to the desktop/notebook space, then GK208 could be like the GT21x series—just low end chip(s).

I'm in the camp that doesn't think there will be a GKxxx chip between the 104 and the 110, and I'm not convinced that a GK114/204 (if it exists) will have a different core count and memory bus width than the GK104. If GK104/114/204 doesn't increase memory speed past 6 Gbps, then I bet that it'll have little or no core clock increases and a fair TDP reduction instead. They'll probably keep one or more Titans for the high end, at least until Maxwell.
 
But at what price point ? Untill they decide to decrese the price of Titan, + the second slimmed GK110, this will put a GTX770 at 550+ $..
nobody will buy a GTX770 with 5-7% more performance of a 680 ( so in the overclocked retail 680 territory) for 550$.

GTX770 = 550$, GTX 780/GK110 = 700$ and Titan 900$ ? Because i dont see them release a cut down version of Titan GTX780 @ 550-600$ for get their 770 at 450$ with the full titan still hovering at 900$ .

I mostly think they will release the slimmed version outside GTX 700 series.. for offer a lower priced Titan, but still keep the card at something around 750 $ .

Well a 320-bit 13 SMX GK110 at slightly lower clocks than Titan will have a significant decrease in performance over Titan. The bandwidth cut + less ROP's + lower clock speed will put a significant dent in performance. I'm also guessing the DP performance will be castrated like on every single previous high-end Geforce card before Titan. I think $599 is a likely price point, with GK114/204 coming in at $449, creating the same $150 price point difference like with gtx580 and 570, and gtx480 and 470.
 
Back
Top