NVIDIA’s current flagship graphics card, the GTX 680, is based on the Kepler GPU architecture. However, being the GK104 variant, it’s not the top model, or “Big Kepler” as it’s been dubbed. That chip has already found its way into the super-expensive K20 Tesla HPC card late last year, with a huge 18700 of those cards finding their way into the Cray Titan supercomputer! However, so far, no GeForce card based on this GPU has been announced and crucially, no announcement from NVIDIA at CES 2013 earlier this month where they unveiled their Project SHIELD portable gaming platform running on Android.
Now, rumors (have pinch of salt at the ready) have leaked out that a variant of this chip is to finally be at the heart of a mega-powerful single-chip graphics card. This will be the GK110 GPU, but unfortunately not the full version, since it will only have 14 SMX units, rather than the full 15 present in the K20 card. This means that it will have 2688 Cuda cores rather than the 2880 of the uncut K20 variant. The GPU will be clocked at a conservative 732MHz and it will feature a massive 6GB of GDDR5 memory with an effective memory clock of 5200MHz attached via a 384-bit bus. Performance is reported to be around 85% that of a GTX 690 and the card will sell for about $900. Finally the name is expected to be called the “GeForce Titan”, fitting in nicely with the name of Cray’s supercomputer.