GTX 680 Transistor count and density revealed

One of two good indicators of performance in a video card often come down to the transistor count and the die size. Generally the more dense the transistor count, the more efficient the chip is and the larger the die size is (within its particular manufacturing process generation i.e 65nm, 40nm, 28nm etc) the more powerful the chip is. Using all the information from a number of sources, including information obtained from the apparently leaked press kit slides from Nvidia, 3D Center have reconstructed a comparative table showing how the most recent powerful GPUs stack up:

What is obvious from this table is just how efficient Nvidia’s GTX 680 GK104 architecture is. With a density of 12million transistors per mm squared it is rougly the same as AMD’s HD 7970. However, with 800 million less transistors and reportedly 10% better performance clearly Nvidia has come up with a very efficient package. The size of 294mm squared is quite small for what is a “high-end” graphics card, in fact its about 40% smaller in size than a GTX 580 and yet it consumes 55W~ less power and gives 20%~ better performance.

What is also apparent is that AMD’s HD 7970/7950 architecture is somewhat inefficient especially when you compare it to its own HD 7870 which has a significant amount more performance per watt with the highest transistor density per mm than any other card on the market. But the most interesting prospect of all is that if the GK104 is only 294mm squared in size then it is not a true performance chip. We expect Nvidia to offer something larger (towards the 400mm squared marker) that should offer even more performance, some are suggesting that will be the GK100.

Source: 3DCenter.org