Nvidia reveals the specifications for the GK110 – full-featured Kepler

Although the Geforce GTX 670, 680 and 690 are among the world’s fastest graphics cards, the GK104 graphics processor is not the full-featured version of the Kepler. Nvidia is working on an even more lavish circuit that will primarily be used for the Tesla computing cards, but which may well appear in the consumer market in the form of Geforce towards the end of the year.

Now the specifications for the mythical GK110 are published, which with 7.1 billion transistors is double up from GK104. The upcoming graphics processor contains as many as 2,880 CUDA cores divided into 15 stream processors (SMX). In addition, the memory bus is widened to 384 bits.

Nvidia also plans to introduce slimmed-down variants with 13 and 14 stream processors, which will reduce the number of CUDA units to 2,496 and 2,688 cores, respectively. However, there is still significantly more GK104, which contains 1,536 CUDA cores.

Finally, some performance figures are revealed, such as that the GK110 should be good for over a teraflop double precision and offer three times higher “performance per watt” than the Fermi-based predecessors. The calculation card Tesla K20 with GK110 will be released sometime during the fourth quarter of the year.