I think it's been consistently shown that bandwidth does not play a large role in ATi's current architecture. The performance difference going from a 2900XT 512MB, to 3870, to 3850 512MB is not very large at all, even though the bandwidth is essentially halved. I think this has something to do with the shader core AA resolve, and hence not taking advantage of the bandwidth because the AA is broken, slow, or if you prefer, 'doesn't perform very well' a lot of the time...something that isn't bandwidth related, nor do I believe going to be fixed with R680, therefore making the 2x256-bit a non-issue if you're comfortable with the current 3870. Hopefully there will be some dedicated logic (or another fix) going forward (R700) alleviating this problem.
On the other hand, nvidia's cards will be well to do with more bandwidth as it ties into their architecture tightly right along with ROPs; something I think we'll see across the whole gambit come the new generation (256-bit low-end, 512-bit high-end, perhaps 384-bit D9EGTS).
Reply With Quote
Bookmarks