There is no such thing as a single universal benchmark that's meaningful in all cases, which is exactly why most such attempts are composites of various application benchmarks.
If you look at video card performance reviews for example, you'll see several different game benchmarks, and if you look closely, you'll see that although there is some pattern, there is no universal rule about performance -- you have to do the specific tests to find the relative standings; often cards interchange positions in different games.
So the biggest starting factor is what games / applications appeal to you, and how do you want to use them (eye candy vs. speed vs. desired resolution). Typically you need a certain threshold of speed in all cases, and you might be able to trade off some of that speed, against resolution first and eye candy second. If you're running LCD, then the fixed resolutoin can be a significant limitation and factor here. Conversely, it also typically stops you at 1280 x 960 or so, and you don't have to worry about 1600 x 1200 performance, etc.
Most games tend to be heavily GPU dependent, some more than others. For most games, an X2 3800 is plenty, and moreover, you can potentially overclock that to compete with the fastest non-overclocked processors (so it had better be fine in this case -- what are the game designers thinking if not so?)
You may be able to squeeze some more FPS with a faster processor, but typically you'd be able to do a lot more with a faster video card. The 6800 base is adequate, and sufficient in many cases, but there's been a lot more development beyond that, and you can get signficantly more performance / eye candy / resolution with faster video cards. Conversely, you can get a lot of actual gaming in with the 6800, if you don't go crazy with the resolutoins, etc. After a certain point, it's pretty much an game of increased resolution / eye candy, as all the cards can do the minminum sufficiently.
Do you need it? Do you want it? Do you want to pay for it?
If you decide you want to pay for it, then AGP vs. PCIe, 7800 vs. 7900 vs, 1900 vs waiting for DirectX 10, etc., etc., are good questions.
Going back to benchmarking, some games conveniently include their own benchmarking stubs. It's great when they do, because you can test the numerous options and tradeoffs directly. For others, you have to find a way to benchmark them; perhaps with 3rd party mods or applications (e.g. FRAPS), or just a seat-of-the-pants judgement based on ongoing gameplay, as most people do.
Facebook
Twitter