Heyyo,
Ok, this one ain't mentioned and it's imperrative! Don't buy any sapphire vidcards with a black pcb! they use some hella-cheap ram that I've had die on me in 3 hours! rediculous eh? rma'd that, got one with the blue pcb.. diddn't think the diff would be big, turns out the blue ones they used samsung ram.. thank god cause my vidcard's fine now. Btw, hint that your vidcard's ram is dying is if you get constant screen corruption as if you were ocing.
Now, someone said tomshardware is biased. For benchmarks yes. but probably cause they don't know that HL2 is ati optimized, and doom3 is nvidia optimized along with someothers. I say you can't trust 3DMark too much either.. both nvidia and ati have been caught cheating in the past on 3DMark. I think far cry's probably the cleanest benchmark they had as it's not geared towards either company. In general, it seems nvidia runs better with AA and AF on, and ati tends to do better when just running crude. Heh, I remember in the past when it was the other way around. Ok, on the pci-e front, nvidia seems to be doing better than ati. Nvidia's thriving on their 6600gt wich I must say brought some faith back to me for nvidia. due to 2 really bad vidcard purchases which I suggest avoiding:
Nvidia Geforce4 MX440 (a geforce2 ti can take it down.. it's pretty much a geforce2 with 64mb of ram..)
and the Nvidia FX5600. When I bought it, the radeon 9600 pro was $20 cad more, and offered 15F/S more... and for a price gap that small? easily worth it. Plus, the fx5600 ran like junk when there was quite a bit of particle effects going.