Really? Hmm, that's very disappointing, and strongly indicates that nVidia has no notebook plans for the 6xx generation.
They do, it's a mix'n'match of both Kepler and Fermi
610 = GF119 (Fermi)
620 = GF117 (Fermi, 28nm shrink)
630 = GF108, GF117 (Fermi, Fermi 28nm shrink)
635 = GF106, GF116 (Fermi, Fermi)
640 LE = GF108, GK107 (Fermi, Kepler)
640 = GK107 (Kepler)
650 = GK107 (Kepler)
660 = GK107 (Kepler)
670 = GF114 (Fermi)
675 = GF114 (Fermi)
And according to some rumors, 680 = GK106
UniversalTruth
Veteran
Yup.
The problem here is that GK106 may end being slower than GF114.
That's just weird. And confusing.
Welcome to the wonderful world of NVIDIA.
I don't think more true words have ever been spoken on this forum.
More funny is, that 28nm and Kepler architecture is not marketed by NV as advantage of 640/650/660M GPUs.
In addition to he backend hole looking pure black, the pcb seems to bend a bit after pcie connector though surely it shouldn't on level surface?
Also, the hole in the backend is differently shaped in left and right ends of it.
And should perspective really turn the backend of the pcb to that much different angle compared to io-panel?
Also, the hole in the backend is differently shaped in left and right ends of it.
And should perspective really turn the backend of the pcb to that much different angle compared to io-panel?
:he problem here is that GK106 may end being slower than GF114.
Maybe on desktop, but not in a laptop power envelope...
There are some rumors about GTX 680M (GK106): http://dell.benyouhui.it168.com/thread-2046420-1-1.html
... P4600@3DM11 and a 256-bit IMC.
This is not far from GTX 560 Ti, although Kepler seem a bit favored by 3DM11.
... P4600@3DM11 and a 256-bit IMC.
This is not far from GTX 560 Ti, although Kepler seem a bit favored by 3DM11.
iMacmatician
Regular
Interesting on the 256-bit memory controller (I guessed 192-bit). Now I'm wondering if they're doing something like what AMD did with Barts, keeping the bus width (of GK104) but using a slower and smaller memory controller.
Also, about the GK106…
The GTX 550 Ti has (116 W)/(170 W) = 68% the TDP of the GTX 560 Ti for (900 MHz)/(822 MHz) = 109% the clock speed. If the relationship between the (highest-end card based on the) GK106 and the (highest-end card based on the) GK104 is similar to that of the (highest-end card based on the) GF116 and the (highest-end card based on the) GF114 (specifically 1/2 the CC count, but if the GK106 indeed has a 256-bit bus it could be more different than I thought), then the GK106 version of the 550 Ti would be around 0.68·(195 W) = 133 W for around a 1.09·(1006 MHz) = 1101 MHz clock speed.
Now the GTX 560M (the highest-end, actually the only, mobile GF116) has (75 W)/(116 W) = 65% the TDP of the GTX 550 Ti for (775 MHz)/(900 MHz) = 86% the clock speed. With the same assumptions as before, a mobile GK106 could be around 0.65·(133 W) = 86 W for around a 0.86·(1101 MHz) = 948 MHz clock speed. Raise the TDP to 100 W and you get (assuming linear scaling) (948 MHz)·(100 W)/(86 W) = 1103 MHz, about equal to a hypothetical 550 Ti-like, i.e. "full," GK106.
Also, about the GK106…
The GTX 550 Ti has (116 W)/(170 W) = 68% the TDP of the GTX 560 Ti for (900 MHz)/(822 MHz) = 109% the clock speed. If the relationship between the (highest-end card based on the) GK106 and the (highest-end card based on the) GK104 is similar to that of the (highest-end card based on the) GF116 and the (highest-end card based on the) GF114 (specifically 1/2 the CC count, but if the GK106 indeed has a 256-bit bus it could be more different than I thought), then the GK106 version of the 550 Ti would be around 0.68·(195 W) = 133 W for around a 1.09·(1006 MHz) = 1101 MHz clock speed.
Now the GTX 560M (the highest-end, actually the only, mobile GF116) has (75 W)/(116 W) = 65% the TDP of the GTX 550 Ti for (775 MHz)/(900 MHz) = 86% the clock speed. With the same assumptions as before, a mobile GK106 could be around 0.65·(133 W) = 86 W for around a 0.86·(1101 MHz) = 948 MHz clock speed. Raise the TDP to 100 W and you get (assuming linear scaling) (948 MHz)·(100 W)/(86 W) = 1103 MHz, about equal to a hypothetical 550 Ti-like, i.e. "full," GK106.
Last edited by a moderator:
iMacmatician
Regular
NVIDIA followed up on their "It's Coming" teaser with an announcement of an announcement at the GeForce LAN / NGF 2012, April 28, 7:30 PM PDT. Their website will live-stream the event.
Interesting. Sounds like a reasonable time to launch the rest of the 6xx family (though I really, really doubt we'll get "big Kepler").
iMacmatician
Regular
"NVIDIA hands out a Crowbar for use in case of Zombies or something else – Another GeForce GTX 690 Teaser?"
They speculate the crowbar is supposed to be used to open the box containing (presumably) the GTX 690.Text on the crowbar said:
Silent_Buddha
Legend
I have a feeling GTC will mostly be unrelated to whatever is announced at LAN.
I imagine that whatever is talked about at GTC will be similar to the Fermi whitepaper announcement. Basically to prep HPC and other GPU compute programmers on the upcoming architecture changes coming in GK110. Basically a 4-5 months heads up so that interested parties can start preparing for any adjustments that need to be made.
Regards,
SB
Jawed
Legend
UniversalTruth
Veteran
+1
This countdown is like they have made a discovery to heal cancer, not about some boring dual-chip videocard launch which will be out of stock almost constantly. So much noise for nothing.
Similar threads
