NVidia's GeForce GTX 600 series debuted last March with GTX 680, based on GK104 chip with 1536 CUDA cores. Based on naming and specifications, it was obvious back then, GK104 wasn't the most powerful Kepler chip NVidia has designed and they didn't deny that. The first products using GK110 were NVidia's Tesla K20 compute cards in late 2012. 18688 of them are used in today's most powerful supercomputer – Cray Titan at Oak Ridge National Laboratory.

NVidia named its GeForce GK110 offering based on the name of supercomputer without assigning it any standard numerical designation.  Today was a paper launch during a small press event, precise benchmarks are expected within few days and retail availability some time later. Even without benchmarks, specifications can tell what a beast GTX Titan is: 2688 CUDA cores, 384-bit memory bus, 6GB GDDR5 at 6GHz, 27cm long, 7.1 billion transistors and 250W TDP. It can also leave a £900 hole in your wallet. At lower resolutions (1920x1080), it should be similar in performance to GTX 690, without SLI issues. At 2560x1600 or surround setups, its memory system shouldn't become a bottleneck, so it will be the most powerful card for those scenarios. Finally, for the big enthusiast with enough money, there are options for GTX Titans in SLI or Tri-SLI, bringing performance that GTX 690 was unable to provide.

Some other features of GTX Titan aren't directly related to graphics performance. It features second version of NVidia's GPU Boost technology, which dynamically adjusts card's clock rate based on load. The main difference is the control of frequency based on temperature reading instead of TDP, leading to higher dynamic frequency. This revision will also appear in future GeForce cards. Double-precision compute was gutted in GK104, making it quite weak at compute, compared to AMD's HD7000 series. In GTX Titan, compute performance isn't compromised, so it will be a best performer in tasks that utilise GPU compute. Finally, just like with GTX 690, NVidia GTX Titan emphasises that it is premium class in its looks. GeForce GTX lit logo on top can be controlled via software to change its colour, either based on temperature or to fit the case. Engadget has a nice gallery showing the card.

All in all, we've got a new single-GPU performance king in unusual circumstances a bit outside the main GPU line. It has much more memory and much larger chip that any other current graphics card and comes at price bracket that was occupied solely by multi-GPU cards before it. Relative to current cards, it might not seem as a bad deal – twice the performance at twice the price. Remembering the 40 nm process based monsters (GTX580, HD6970) however, we might have had this 28 nm process based monster at lower price point if NVidia wasn't the only one offering it.