hannibal :
Hard to say, but there is not so big difference in clock speeds. The faster memory is good when using high resolutions and high AA.
If that's the case, then 22.4GB/sec would've always remained fine for those of us at 1024x768... But it's not, as games get more and more demanding. So the assertion that drastically higher memory bandwidth won't make a difference is utter BS. Remember, while increased resolutions do increase how much writing the card needs to do, then again, newer games with more shaders require continuously more passes (overdraw) in order to create each frame. So if you consider that as a third dimension, even at 640x480
Crysis has a higher framebuffer-related bandwidth load than any supported resolution for, say,
Jedi Knight or
Half-Life.
It's worth noting that the RV770 is massive leaps and bounds over RV670, in terms of both texturing and shader power; we go from the 3870 with its 16 TMUs and 320 stream processors at 775MHz, to the 4850 with 40 TMUs and 800 stream processors at 625MHz; that's a 101.6% increase for both, yet the card does NOT perform twice as well as the 3870. Clearly, it is being bottlenecked by the memory bandwidth available in the games being used; a new memory architecture helps, but it hardly can cover the fact that in fact, the 4850 has slightly LESS memory bandwidth than the 3870, dropping down 11.1% to go from 72.0GB/sec to 64.0GB/sec.