14 DAY TRIAL // One would think that, as far as NVIDIA is concerned, people wouldn't be looking for info on anything but the GeForce GTX 680, but that is not the case at all.
The folks at 3DCenter.org were actually able to get some information on a certain graphics processing unit that would come later this year.
Then again, that is not a totally correct way of saying it, since the specs have been “approximated” rather than discovered, which means that some speculation was involved.
At any rate, 3DCenter.org posted what are believed to be the specifications of the Kepler-based, 28nm GK110 chip.
Indeed, not GK100 but GK110, which means that this is probably the strongest of this series, or will be.
The die size is bound to be of around 550 square millimeters, 87% larger than that of the GK104, which can only mean big add-in cards.
Additionally, the SMX (streaming multiprocessor 10) design may go through some modifications to the CUDA cores and memory bandwidth.
NVIDIA may choose to increase the GDDR5 memory interface to 512 bits, while not pushing the CUDA cores to work as fast as before, or even adding that many.
That is to say, the CUDA core count is of 2,000 to 2,500 (not that big an increase over GK104), which means the overall increase in compute power is of just 30% or so.
On the other hand, the transistor count will be very large: 6 billion. For comparison, the GK104 is said to have only 3.54 billion (the die area is of just 294 square millimeters after all).
Finally, the video boards powered by it should bear the name GeForce GTX 700 series and will run on 250 to 300 W of energy.
If the Santa Clara, California-based company launches the GK110 this year (2012), it won't be before August. As for the GK100, it probably still exists but we haven't heard anything new about it lately.