Performance can't accurately be predicted at this time, alas

Mar 6, 2012 09:34 GMT  ·  By

Though the official launch of the GeForce GTX 680 is rumored to be scheduled for March 23, it is said that the Santa Clara, California-based company will speak of the board a bit sooner.

According to VR-Zone, NVIDIA will reveal some details, in official capacity, on March 12 this year (2012).

That is less than a week from now, so this calls for a tally of what is known about the product so far.

There will be 1,536 stream processors in the GK104 graphics processing unit, though NVIDIA calls them CUDA cores.

Speaking of which, the shader hotclock will be in effect: the SPs will run faster than the GPU, despite what rumors have been saying.

To elaborate, the graphics chip will operate at 706 MHz and the CUDA cores at 1,411 MHz.

The memory controller will exhibit some improvements as well, actually pushing the VRAM to a 6 GHz performance.

As such, the memory bandwidth will be of 192 GB/s, the same as the one on the GTX 580, despite the memory bus being narrower.

NVIDIA previously said that the Kepler would be unbeatable when it finally arrives, and what we have seen so far shows that it is at least possible to live up to that boast.

Moving on, the high-end video controller possesses two SLI connectors, leading to 3-way SLI multi-GPU support. The weird power plugs may interfere with this capability though.

Finally, there are two DVI video outputs available, plus an HDMI and a DisplayPort. Not exactly the same multi-display support as AMD's Eyefinity, but 4 monitors is still a large number.

We continue to hope that the experimental, stacked PCI Express power connectors don't make it into the final incarnation of the Kepler. Perhaps NVIDIA will say something about this at the upcoming event. If not, we'll just have to wait and see.