SMOOTHVISION 2.1 tested
Testing the efficiency of F-buffer addition to the Radeon 9800 Pro is problematic, as we need incredibly complicated shaders to ensure that the present Radeon 9700 Pro will, in fact, resort to needing to multi-pass the shader in question. No present games include pixel shaders that can fully stress the pixel shader length limit of the present card. We'd need some custom code that goes beyond the specifications of DirectX9.
We can, however, attempt to test whether ATI's SMOOTHVISION 2.1 has any real merit to it. Remember, ATi claimed that the newer form of SMOOTHVISION increased the performance of the 9800 Pro, relative to the 9700 Pro when higher levels of anti-aliasing and anisotropic filtering were applied. Rather begrudgingly, I managed to overclock our 9700 Pro to 9800 Pro levels (380/680). Then, benchmarking Unreal Tournament 2003 Build 2206, we got the following results at 1280x960x32 with no anti-aliasing or anisotropic filtering applied.
The 9800 Pro is obviously faster than a stock 9700 Pro. What's interesting is that the overclocked 9700 Pro performs almost exactly like a standard 9800 Pro, given that they both use the same basic speeds. Now let's bump it up to 4x AA and 8x AF.
Here, although the overclocked 9700 Pro is running the same basic speed as the 9800 Pro, it does fall around 10% behind. This lends weight to the improved SMOOTHVISION performance once reasonable levels of AA and AF are applied. I must temper this by stating both cards were running different drivers. The 9800 Pro ran the 6307 set, whilst the 9700 Pro ran the 6292 set. Both cards steadfastly refused to run each others' drivers due to inf errors. I'd like to think that it's less to do with drivers and more to do with a greater efficiency with respect to image enhancement.
