http://www.3dcenter.org/news/die-akt...rformance-chip
http://www.3dcenter.org/news/nvidia-...radeon-hd-7970
Printable View
Interesting, TPU has posted it with that article as a source, calling the rumor "reliable"... http://www.techpowerup.com/160263/NV...Machinery.htmlQuote:
Originally Posted by Andrea deluxe
?_? <-- my face after following this thread, I've never seen so much rumored specs floating around that I probably wouldn't even notice if Nvidia confirmed them and call them fake. :P
^QFT
This thread's reliability is inversely proportional to its number of pages; and directly proportional to the numbersss of countlessss-ish rumorsssss...
I was hoping at-least few infos were solid at this point, hell even the cards' numbering ain't confirmed; no one knows if it will still have "Geforce" on it. LOL!
Might be time for the mods to move this thread.:D
But that would ruin all the fun! :p:Quote:
Originally Posted by labs23
i think the lack of info is because they will launch a mid range card below the current 580 price. and if they told us that then they would loose all sales from when they said it, until it launches. if they had something really powerful coming soon, they would be talking about it to slow down 7970 sales.
Yeah, it is crazy, never seen that in 10years, i think this is a record... not one day pass without a rumor about kepler ... Something strange is behind it, and i dont know what to think about it....Quote:
Originally Posted by RPGWiZaRD
For me the only way it seems to make sense is that maybe their big X80 chip is delayed somewhat. They said stuff about 7970 not impressing bla bla bla and thats all, they havent touched on when they are going to have something that will make it look ordinary. Now rumors of a X60 chip are pretty consistent so im gonna go with that. Its going to be a ~350$ chip seems to be the most sensible assumption I can make out of it.
Maybe this is like Fermi all over again but this time it might be worse because they cant release a X80 chip so we are going to see something like GTX465 ? Which in the short term will be a disaster till they get their X80 and X70 ready with decent thermal performance.
Maybe March will bring a new flavour... Nvidia should have something out by then, and AMD will have New Zealand out
edit** OMG 777 post! better go to the casino and bet it all on red (they don't offer green as an option)
LOL :). Depends on what casino you go to ;).Quote:
Originally Posted by kuroikenshi
It might be worse? Not really. It looks much better. With Fermi they needed 6 Month form Amds High-End and Performance chip to launch and all the small stuff was also 6months late compared to amds small stuff. This time they'll launch all chips except of the top-dog in April (At least in Notebooks). That's just 1,2 Month behind Amds smaller parts. Nvidia Top-Dog might again come 6 Month late, but probably you just can't launch such a big chip with early process.Quote:
Originally Posted by Zloyd
You can't bet on 0 or 00?Quote:
Originally Posted by kuroikenshi
Real NVIDIA Kepler, GK104, GeForce GTX "670/680" Specs Leak Out
http://www.brightsideofnews.com/Data..._Mockup_68.jpgQuote:
In the past few weeks, we've seen various fishy rumors on the product specifications of first discrete GPU using the upcoming 28nm Kepler architecture the GK104. While we have known parts of the specifications, such as no hot clocks, the doubling of Streaming Multiprocessor (SM) node from 48 to 96 CUDA cores (i.e. Stream Processors), 256-bit memory controller, the real specifications are (finally) here... even though, our information differes minimally from information originally posted on 3DCenter.org.
Quote:
NVIDIA Kepler GK104 Architectural overview: at first look, very similar to GF110, but then you take a deeper look: 1536 Stream Processors instead of 512!
So explain to me again what makes their leak "real"?
-PB
Liked this bit:
Finally, just about time... Just a hint on "when" so maybe that April news is true after all, April 1? XDQuote:
You won't need to wait for too long, as NVIDIA is already starting pre-sale activities, and getting ready to counter AMD and their momentum with the Radeon 7700 (Cape Verde, February 15), 7800 (Pitcairn, March 6) and 7900 Series (released).
1536 really? uh i am shocked
:D
I believe it when I see it.
I would love it to be true though, too many suggested specs around at this point. If it's true Charlie's positive article regarding GK104 would turn out to be true and Nvidia indeed would truly have a winner in their hands. I dunno but it just seem unrealisticly positive regarding performance/price haha.
96 cores with 4 Schedulers ??? i guess it is SuperScalar ??Quote:
Originally Posted by cold2010
48 ROPs with 4x64-bit (256) memory controllers ??? really ???
This is a Fermi pic with 96 cores per SM
I say FAKE ;)
http://img703.imageshack.us/img703/8...nstitrexxw.png
I have put side to side, the BSN scheme and the GF110 ones.. Just for see the difference in density of the SP ( ofc this is a schema lol, but this give a idea of it )... now try imagine it in a 340mm2 cores instead of 550mm2... I dont even want try to imagine a GK100-110. this make 3x more SP in an SM...
The only way is barely divide the cuda cores (SP ) size by 300%, or i dont see how they can do it in 340mm2.
Quote:
Originally Posted by Aten-Ra
:rolleyes:Quote:
... As you can see in our architectural mockup ...
Lol yes, i doubt we will have a core schema before the release or before the " first presentation" of the card by Nvidia general..
ok, i didn't read the article. I thought that was supposed to be a legit pic of the architecture and commented on that assumption ;)
I have to assume using hotclocks decreases transistor density but who knows by how much.Quote:
Originally Posted by Lanek
What was the most recent Nvidia hotclocked GPU that we have a real die shot of?
Shaders, for AMD, don't take up as much of the die as one would imagine, at least in previous vliw5 architectures. Jawed at b3d did a breakdown of Juniper, I think, few years back. I'll see if I can find it.
Nope, the GF1X4 went superscalar to have two schedulers issue up to 4 instructions. Here you have four schedulers to issue the same 4 instructions.Quote:
Originally Posted by Aten-Ra
Oh and I count 8x8 = 64 ROPs. :P Besides, ROPs don't have to be coupled to mem controllers, see GCN.
Well these guys surely seem the most confident out of the rumors I've read, but I still don't like how these numbers look eerily similar to AMD's design. I guess the simple answer would be that Nvidia decided their superscalar design was too expensive and resulted in poor yields compared to AMD, but even if they did triple the core count, if they removed the hotclock wouldn't that result in just a minimal increase?