NVIDIA GeForce GTX 780 To Be Based on GK114 GPU

That's why I said superior

And why I thought 70% was rather fair. Sure you can see 100% scaling.
But you can have poor or even negative scaling, you will be depending on profiles, and despite Frame Metering there are latency as well as MS issues.

13, 14 or 15 SMX, even at 35% lower speed, all housed in a single die.
Yeah I'll take that over 2x 7 SMX talking to each other which one should go first.

In the end it will all boil down to TDP, what NV/TSMC make of it., and what TDP target NV sets for GTX 780.

 

I'll take 30 SMX talking to each other

 

It might not even make it into the Tesla cards (and given that the K20 has 5 GB, it seems even less promising).

 

Got a link? 5GB doesn't make sense, unless the bus is actually 512b.

 

um 5gb would make no sense for a 512 bit bus. the K20 specs indicate a 320 bit(384 with one controller disabled) bus so yes 5gb would make sense.

 

Oops, my bad.

 

If a 20% increase is all it gives thats fine. Its not like there are any games a gtx 670 cant handle nowadays, let alone gtx 680.

 

https://bongous.com/

 

If we are still talking the GK110, then I think it's +/- 40% depending on final clocks, at least that the last credible number I've read. I think @ beyond3d.com they are speculating clocks in the 700+ MHz range for the K20. IF it came out as a consumer card, I imagine the clocks would be a bit higher. Much more than that in compute - I think it would be a monster F@H card.

 

http://translate.google.com/transla...&eotf=1&u=http://www.toldpriser.dk/toldregler

They control (the control adds another tax) quite vigerously...customs are not relaxed.

 

Still curious about hardware frame metering though! nVidia is still tight lipped about this.

 

Well if maxwell's GM104 increases in performance over GK104 like GK104 did over GF114, we're looking at comfortably outperforming GK110. gtx680 is at least 70% faster at 1920x1200 and 89% faster at 2560x1600 than gtx560ti.

 

GTX680 is nearly twice as fast as the GTX560TI:
http://www.computerbase.de/artikel/grafikkarten/2012/test-nvidia-geforce-gtx-650-ti/6/

That is a great result because the GTX680 has only 33% more pixel/rop performance and 50% more bandwidth.

We can expect the same from the GK110 over GF110.

 

LOL, yeah I know. I am leaning more towards anywhere from 2.5GB, 3GB, 5GB or 6GB depending on rumored 320 or 384 bit bus. When I said up to 24GB, I'm just going by what Tesla can offer so don't take that as literally transferring over to a consumer GPU. Would be sick though, wouldn't it?

 

Only if windows could really leverage it, no more slow DDR3!

Windows 8 seems to be loading up vram /w browsers though maybe 7 did that as well I just never noticed.

 

Well if this bs rumor is really true, which i doubt very much, i still think it will be a GK110 chip. Then im going AMD 8900 for sure, screw NV and this poor midrange/lower quality chip., GX 110 is where is at, not GX 104/114.

104/114 is the troublesome mainstream buggy chip., i know i've been following NV drivers for years and i saw a certain pattern by these chips or better yet lower-end mainstream chips, these have the most bugs..Including full GK104 aka 680 gtx.

 

Same here, I'd be a bit let down by NV too, leaving us gamers out with only a GK114.. It'd be like Intel slacking off with only a quad-core CPU ever since a 6-core Gulftown 2 generations ago.

 

Agree. The 680 is 35-40% faster than the 580. Even if the 780 is only 35% faster than the 680, that would still be roughly 90% faster than the 580.

 

Ironically, if GK110 is 30% faster than gtx680, and the last round of rumors regarding the hd8970's performance are true, we are looking at the same performance delta between the gtx500 and hd6000 series and the upcoming refreshed parts.

 

Could very well be, yes. But personally I suspect the 780 is rather 40-50% faster than the 680 due to higher bandwidth. I believe that could translate 1:1 into more performance. Big question is what AMD will do. 15%, 30%, and compared to what? 7970 or 7970 GE?

But it all also depends on Nvidia. 10% more performance might be enough for them to improve yields and not go overboard with power consumption.

 

I was referring to the hd7970GE. 15% faster than the hd7970 for any top of the line refresh from either company would be really, really, really disappointing.

I guess it's going to come down to core clock speeds and what they can squeeze in under whatever TDP they are comfortable coming to market with. I hope (at the least) they'll unlock the voltage this time - although I doubt it. With Nvidia done supplying ORNL, they are obviously ramping up GK110 pretty heavily. Hopefully we'll get some Geforce based information/rumors soon. I'd happily take a late December paper launch, something I see Nvidia doing to reclaim the undisputed single GPU performance crown (like what AMD did last year with the hd7970).

 

People shouldn't get to excited after all the hype from past years and how people were sorely disappointed. 10-20% is likely unless they had improved on this issue http://www.realworldtech.com/kepler-brief/