Nvidia GeForce GTX 680
When Nvidia’s Fermi cards launched amidst much hype and fanfare, it was a move forward from the 400 series, but not nearly as far as many had hoped. In hindsight, perhaps Fermi was more of a stepping stone to Kepler, not unlike Windows Vista to Win 7. As we’ll see, Kepler is a significant improvement in hardware technology, performance, features, and software across the board.
Below is an image of the Kepler block diagram. It actually doesn’t look too dissimilar from Fermi on the surface, and the diagram illustrates 8 SMX, 1536 CUDA cores, 8 Geometry Units, 4 Raster Units, 128 Texture Units, 32 ROP units, and a 256-bit GDDR5 bus.
Below is a detail of the SMX, with each cluster containing 192 CUDA cores, 16 Texture Units, and Polymorph Engine 2.0.
Kepler certainly brings performance improvements, but the 28nm manufacturing process also reduces heat output from about 880 BTUs found Fermi’s GTX 580 to 660 BTUs now in the GTX 680. However, perhaps the most significant improvement, is that Kepler brings twice the Performance Per Watt when compared to Fermi. In a word, that’s nothing short of awesome. Lower power consumption appears to be a hallmark of Kepler, and that’s particularly encouraging. Killer performance is one thing, but killer performance at considerably lower power consumption is something special.