The new 'Titan' graphics card will sport the GK110 GPU, a GPU previously only used in Nvidia's Tesla K20X cards. The specifications are impressive: The GPU contains 2880 CUDA-cores, of which a generous 2688 are enabled. The reason why not all of the CUDA-cores will be enabled is likely because of a binning process; there will be few chips on which 100 percent of the cores will work as they are specified to. The clock speed will also be set at a fairly low 732MHz, however, as a result the card will have a TDP of just 235W. In comparison, the GTX 680's GK104 GPU comes packed with just 1536 CUDA-cores, but has a much higher clock speed at just over 1GHz. Beyond the immense number of CUDA-cores, the chip will have a 384-bit memory bus, along with a shocking 6GB of GDDR5 memory that will run at 5.2GHz.
The performance is expected to be roughly 85 percent that of the GTX 690, but remember, this card will only have one GPU onboard, not two.
Seeing as Nvidia already has the 670, 680, and 690 in operation, and there is no numerical name that the card can be assigned within the current generation, it makes sense that Nvidia has decided to name the card 'Titan'. Rumors say that the name 'Titan' may be inspired by the Cray Titan supercomputer, which contains nearly 19 thousand Nvidia Tesla K20X cards.
Availability is rumored to be around the end of February with a hefty price tag of $899.