lehpron

Whenever you find specs online, you have to analyze it for realism based on what has been done.  Don't put in faith without investigating.

Stock GTX780's 2304-core base clock was 863MHz, boost of 900MHz; stock GTX Titan's 2688-core base clock was 832MHz, boost 876MHz. In all probability, a 2496-core GeForce single GPU would have base and boost in between, 847MHz and 888MHz, respectively.

All nVidia dual-GPU graphics cards (except the first one, GeForce 7800GX2, which was an OEM experiment made just for Alienware in winter of 2005) had lower frequencies than the single GPU variant just to reduce power consumption and maintain competitive advantage of getting two single-GPU cards in SLI.  Both GPU and Vram share power circuitry, that why the dual-GPU card isn't a double of one card's power.

nVidia won't cannibalize their products-- except if competition tries to steal customers or can't react.

Considering the fact that only because GTX680 used a GK104 and made it a 195W card, that GTX690 became a 300W card with a slight down clock (10% downclock for the pair allowed for a 53% power increase).  Compare that with GTX580 which was a 244W card with a GF110 at shader frequency of 1562MHz against the GTX590 with a pair of GF110 downclocked to 1242MHz just to keep it at 365W (25% downclock for the pair allowed for a 49% power increase)-- this is the scenario that a GTX790 is looking at since each single-GK110 graphics card is in the range of GTX580's power.

Meaning, GTX790's base clock may be as low as 700MHz just to keep the total power of the card in the 300-350W range and maintain the appeal of still getting a pair of 780 or Titan in terms of performance-- unless nVidia felt AMD would threaten their cash cows with their upcoming next gen cards, there is no way the base clock would be above 900MHz.

FYI, Tesla K20 also has 2496-CUDA cores and a reference base clock of 706MHz with 320-bit 5GB Vram rated as a 225W card.  If we used the 680/690 logic on the Tesla card, then the dual-GK110 version could be 642MHz base drawing 346W.  But that assumes a pair of 320-bit 5GB, not a pair of 384-bit 6GB.  So either way the power draw will be high despite the lower base clock.

nVidia could always shrink the GK110 die, then similar from GTX280's 65nm G200 at 236W to GTX295's pair of 55nm G200b's put it at 289W, then the dual-GPU card would only add 25% power because the dies were shrunk, thus GTX790 could be a 312W card-- if nVidia shrunk the dies...

(source for all "Maximum graphics card power" numbers are available from GeForce.com, pick a card then go to specifications section)

Ahson

GPU MODEL: 2x Kepler GK110
CUDA CORES: 4992
GPU BASE CLOCK:~ 1100 MHz
GPU BOOST CLOCK:~ 1150 MHz
MEMORY CLOCK:~ 1500 MHz
EFFECTIVE MEMORY CLOCK:~ 6000 MHz
MEMORY SIZE:6144 MB
MEMORY TYPE:GDDR5
MEMORY BUS TYPE:768-bit
MEMORY BANDWIDTH:– 404 GB/sec

PCI-E 3.0 x16
DIRECTX SUPPORT:11.1
OPENGL SUPPORT:4.2