Originally Posted by derickwm
That short PCB cracks me up everytime.
Same! It reminds me of a 550ti sort of thing!
EDIT: The short PCB on the Galaxy card someone linked is stupid? I mean why make the card longer then it has to be.. Especially if it's a blower style cooler.. Makes no sense, wouldn't making it smaller generate more sales? (Some cases won't fit that).
As for the MSI Reference card, it looks nice to me. No "cool" stickers and stuff on there, this is why I don't buy reference partially.. As someone mentioned the fake nuts look like crap.
Originally Posted by brettjv
Now this theory makes some sense, as there's definitely historical precedence for something similar (i.e. the Gx1x4 chips being used at the x60 card tier ... see the 460/560 in Fermi). However, that's really not 'why' we see so little difference. The 'why' is simply that the two cards are identical aside from 14% more CUDA cores on the GTX680. All the previous 2nd tier cards had gimped memory subsystems (less ROP's, less bus width, less vram) as well as fewer cores.
If anything, the 'why' ... is that nV has copied AMD's strategy for the 2nd tier card ... one that AMD have used for all three DX11 card generations (5850/6950/7950). In all cases, the 2nd tier card is the same architecturally as it's big brother, aside from a disabled cluster.
I think it's fairly obvious that unfortunately nV weren't able to get GK100 (which would've been our 670/680 chip in this first Kepler generation) working in time, and they discovered that the GK104 at high enough clocks was sufficient to challenge AMD 79xx-series anyways, so they converted GK104 to the top-tier chip. That much is pretty much 'known' at this point, actually.
But the espoused notion (elsewhere on the thread) here that 'GK110 was supposed to be the 680 and GK104 was supposed to be the 670' makes no sense to me as a theory.
GK110 should be the next generation top-end chip, i.e. GTX770/780 (just as GF110 was the 2nd gen i.e. 570/580 in Fermi). However I'd not be surprised if it wasn't out as Tesla product first
Yes, I agree with you on the Tesla product theory. I do think that nVidia are copying AMD's idea here (not gimping the memory) on the lower models. (It would indeed making drivers easier I bet) - To not have to account for all different sorts of memory limitations.
As for the GK110, from what I've read it was ready and almost complete. But nVidia compared their GK104 to the 7970 and such. So they just released it as a GTX680 now so AMD didn't hold the market for too long. They'll have more sales this way, and also that the GK104 would have been cheaper to produce. Which means bigger profit gains on each PCB sold (for the companies at least).
GK110 is going to be amazing, I wouldn't be surprised if we got it before the next gen of cards. If nVidia made a GTX685 (like the old 2xx series) which wasn't GTX560, 560Ti, 570, 580. But GTX260, GTX275, GTX280, GTX285, GTX295. I'm just speculating here, we may see a GTX780 launched before AMD's next flagship and nVidia may be making their money here.
All in all, I do think that nVidia have "working" GK110's under NDA that could have been sold, but they've just made less work for themselves by marketing a 660ti as a GTX680, which also means more profits for everyone except the end user.Edited by Matt-Matt - 5/8/12 at 11:57pm