Back in March NVIDIA announced three new Graphics Processing Units (GPUs) - the GeForce 7900 GTX, 7900 GT, and 7600 GT. The GeForce 7900 GTX became NVIDIA's flagship GPU by replacing the GeForce 7800 GTX 512MB. A much higher-clocked 512MB version of the GeForce 7800 GTX offered a 30-40% increase in performance over the 256MB model and competed against ATI's X1900XT.
The GeForce 7900 GTX is manufactured using a 90nm fabrication process. The die shrink from the 110nm GeForce 7800 GTX 512MB made it necessary to redesign circuitry. During the redesign, NVIDIA's engineers implemented optimizations that allowed the GPU to be produced more efficiently. Even so, the GeForce 7800 GTX 512MB was outfitted with ultra-fast Samsung 900MHz GDDR3 memory and ranks near the top of "NVIDIA's best-ever GPU" list.
The GeForce 7900 GTX was another hard launch by NVIDIA, meaning that graphics cards were available to purchase on the day of the launch. The GeForce 7800 GTX 512MB was also a hard launch, for a few hours at least, but demand quickly exceeded supply, availability became scarce, and the price skyrocketed.
Having majored in Economics a really long time ago, I learned that taking a step backward is undesirable. Here is how that applies to SLI. Once you're sold on SLI technology, you'll probably continue to use SLI with future GPUs. Returning to a single GPU is like, well, taking a step backward.
Being able to use a second graphics card from a different add-in manufacturer was made possible with the Release 80 Driver. With cross-card compatibility, SLI no longer requires that graphics cards be identical, but they must still have the same core GPU.