While the numbers for the G80 were impressive in our existing test platform, the new setup shows much better performance gains for Nvidia's DX10 based cards. The DX9 hardware gains somewhat, but there are fewer gains due to the nature of the limitations in what that hardware can do.
That said, there are a few things we would like to point out. The first is that Nvidia's cards had better gains from a faster CPU and platform. Part of this can be directly attributed to the driver; Nvidia's driver is known to utilize more of the CPU, so the more the CPU can deliver, the better the performance.
Along the same vein, ATI had fewer gains, but this is praise for Terry Makedon and the Catalyst driver team. It shows that an ATI card will not be as affected by slower components over different systems. This is true to a point anyway; once the system becomes the bottleneck, it doesn't matter what you put in the card, as not much will happen.
One thing is for certain: you need bleeding edge horsepower to get the most out of DX0 hardware. If you are in the market for a high end card, you'd do best with an entirely new system.
We upgraded our system to make sure we'd get the most objective results for our graphics reviews. The system might be upgraded once again over the next two quarters, but only time will tell if this will be necessary. For now the choice is clear: Conroe is the best choice for testing.
The long and short of this experiment is that you need a high speed platform to get the most out of the new DX10 hardware. If you were planning on getting a $600 graphics card to replace your 1-year-old graphics card, it would behoove you to rebuild your box. Of course, this means that the whole graphics upgrade will cost you a lot more than just the graphics card.
If you don't do the job properly, the net effect will be like hooking up a pair of garbage speakers to a Bose or Klipsch sound system. The effect would be the same... less than optimal performance, and an experience that is far from ideal given the money you spent.