Arithmetic Power (Tests)
To do that, synthetic tests using procedural textures (which are calculation-intensive) are generally indicated. Of course, Nvidia likes 3DMark Vantage and its Perlin Noise test, with which we measured a performance increase of 129% between the 9800 GTX and the GTX 280. But, given the importance that Nvidia places on that test and the ease with which new drivers can be optimized (differently for the GTX 280 and the 9800 GTX, by the way) to make the test do what one wants it to do (see our article), let’s look first at the results with a forgotten version of RightMark 3D with Pixel Shader 2.0 (Direct3D 9.0). Since the values from the different tests are highly variable, we’ve expressed the results in terms of performance differences to be able to show them on a single chart, with the 9800 GTX as reference.
The improvement is much more modest, though it’s undeniably there – in fact it’s slighter for the procedural shaders and much more evident with complex lighting effects, with up to a 78% improvement. Now let’s move to Version 2 of RightMark and its arithmetic shaders 4.0 (Direct3D 10.0).
Here again the gains, while in evidence, are closer to the increase in the floating-point-TO-textures ratio than the figures produced by 3DMark Vantage.