It will use the GK110, just like the Tesla K20 compute cardPeople might think that NVIDIA is done releasing graphics adapters and/or compute cards based on the GL110 graphics processing units, but they would be wrong according to a new report.
Arab PC World has posted information about an alleged Quadro K6000 adapter, which is basically a GeForce GTX Titan for professional graphics designers.
Some of the compute units and memory controllers will be deactivated on the GK110, but the Quadro K6000 should still have 2,496 CUDA stream processors (13 SMX) and a 320-bit memory bus.
Currently, high-end workstations can use Quadro 6000 adapters, based on the Fermi architecture. The Quadro K6000 would finally replace said architecture with Kepler.
It would be about time, since low-end and mid-range Kepler Quadro cards are already out, as we thoroughly covered here.
In addition to standard performance, the Quadro K6000 will be better than the Quadro 6000 and Quadro K5000 through a higher number of CUDA cores and geometry engines, plus 1TFLOPS of double-precision performance, double what the 6000 model allows for.
That is all we know right now unfortunately. The actual specifications will only surface when NVIDIA finally launches the K6000, assuming leaks don't expose the clock speeds before that.
As for the price, somewhere between $2,500 / 1,921-2,500 Euro and $4,000 / 3,074-4,000 Euro is our best guess.
In addition to the GeForce GTX Titan, the Tesla K20 is powered by the GK110 graphics processing unit, but that card is a compute accelerator for supercomputers, not an actual video board. Taking that into account, one might say that the making of the Quadro K6000 was a necessary move on the part of the Santa Clara, California-based company.
Sadly, we do now know when the release will be. “Sometime in the future” is all we've got, and NVIDIA refused to comment on the story.