'Nvidia GTX 660Ti wordt afgeslankte GTX 670 en komt half augustus'

Looking at the data for AMD, it is clear that memory bandwidth plays a substantial role in graphics performance. We identified 3 pairs of graphics cards with similar GFLOP/s, but very different 3DMark scores. In each example, one of the two graphics cards has ~2X the memory bandwidth of the other and higher performance as a result. The Radeon HD 3870 and 4670 were the pair we mentioned on the earlier page. The 3870 has 2.13X the memory bandwidth of the latter, which translates into the 36% better performance that we already observed. In a similar vein, the Radeon 4870 and 4850 achieve 14% and 27% higher 3DMark scores over their bandwidth starved cousins. The 4870 actually understates the advantage of memory bandwidth, because the shaders are less capable than the 5850’s – about 12% lower in terms of GFLOP/s. All together, the 78% higher memory bandwidth raises the performance by 30%.
....

What is intriguing about the examples above is that both AMD and Nvidia graphics cards show similar sensitivity to the memory interface. In most of the cases we analyzed, 2X higher memory bandwidth yielded ~30% better 3DMark Vantage GPU performance. A good estimate is that performance scales with the cube root of memory bandwidth, as long the memory/computation balance is roughly intact.

[Reactie gewijzigd door sdk1985 op zaterdag 28 juli 2012 02:00]