ATI and NV seem to handle memory differently. Comparing their hardware on just memory, when there are other variables, just isn't going to work. When the 1GB 4870 comes out, compare that to the 512MB version, and then we'll see if there's a 512MB limitation.
After using a significant amount of SLI configurations with 512 Megs, 768, 896 and 1 gigabyte. The higher the memory. The more consistent the experience.
I'm sorry Chris, but I have to disagree with this statement. If it were the same hardware and different configs, then yes, I could see this being a valid test. But comparing a 512MB 9800GTX to a 1GB GTX280, and saying the framerate differences are ONLY due to memory, is complete bull. There are numerous other hardware differences that affect it. Take two cards that are the same hardware, and compare them using two different memory configurations. Right now, the only high end single GPU we can do this with is the HD4870, well, once the 1GB version is released.