The same poster said that 470 is slower than the 480 by 8%. What a joke. GTX480 is around 25-30% faster than the GTX470. Even overclocked to 800mhz (i.e.,
32%), the GTX470 is still barely as fast as the GTX480.
GTX470 is about 20% faster than GTX460 1GB (stock reference clocks). However, GTX460 has higher texture fill-rate. Depending on the game, this difference may shrink (Civilization 5) or increase (Metro 2033 with Tessellation). At $220 I'd pick the 470 myself over the 5850, but some GTX460 1GBs can be found for $170, still making them a better price/performance mid-range card at this time.
I think there is confusion here. You can't measure things linearly when it comes to SLI/CF. For example, if you take an HD5770 and double its theoretical performance, it will be faster than a GTX470. However, CF doesn't scale 100%. We know that a single GTX470 is about as fast as dual HD5770s in the real world. But can you claim that a single GTX470 is 2x faster than a single HD5770? Not entirely correct imo. (That's the flaw in the argument which assumes when you CF 2x 5770s, they have perfect scalability).
Similarly, if one considers an overclocked GTX470 @ 800mhz, it will come pretty close to GTX480 in speeds. Well a
GTX480 is often as fast as a GTX460 768/1GB in SLI (
again due to less than 100% efficiency of SLI), but generally loses by 20% to SLi 460 1GB. Someone may be able to make claims such as
"an overclocked 800mhz GTX470 is 80% of the stock GTX460 SLI setup". This is what that poster really wanted to say I believe (which is not the same as saying a stock GTX470 is 2x faster than a single GTX460).
Of course a GTX460 can also be overclocked, and beat a GTX470 (check out the latest Xbitlabs article on Palit overclocking). Thus, comparing a GTX470 overclocked to GTX480 speed to a stock GTX460 SLI setup is also misleading
