There’s a pleasant surprise with the ROPs, too – they’ve increased in number from 24 on the G80 (16 on the G92) to 32. To be sure that they’re completely used, Nvidia has now introduced a 512-bit bus, which will surely be, given the characteristics of this GPU, more useful than the one on the R600.
With a 78% gain between the GTX 280 and the 9800 GTX, we’re exactly in line with the theoretical figure, since the doubling of the ROPs is impacted by the drop in frequency (675 MHz on the 9800 GTX).
And since we’re on the subject of AMD, we should note that it’s becoming more and more urgent for them to review their high-end GPU, still desperately clinging to 16 texture units and 16 ROPs since the X800 in 2004! Whereas AMD’s GPUs remain competitive on the level of raw processing power, on the other points they’re beginning to be seriously outdistanced by Nvidia, who’ve been making improvements with each new generation. Let’s hope that the new AMD architecture, which should be making its appearance soon, will close the gap.
That said, the results for the HD 3870 X2 are an opportunity to point out that while ATI is outdistanced in the number of units per chip, don’t forget that the manufacturer’s new strategy is to put bi-GPU cards up against single-GPU cards from Nvidia! That point, plus a GPU frequency that’s still higher (825 MHz) puts the 3870 X2 in front on this test, which is highly synthetic but valid.