Conclusion
With the new DirectX 9 mainstream and entry level cards, NVIDIA introduces the new technology to the mass market. Actually, it's a positive trend, but it is overshadowed by the fact that the new features cannot really be put to use by anyone due to a lack of appropriate games. What counts in the end is performance in the current games - and in this respect, it doesn't look particularly good for the new cards.
The GeForceFX 5600 soundly beats the established Radeon 9500 (non-PRO) cards with 64 MB and 128 MB. Compared to the Radeon 9500 PRO, however, it mostly draws the shorter end of the stick. In view of the pixel pipes, which have been reduced from eight to four, it still remains to be seen whether the Radeon 9600 PRO can beat the "old" 9500 PRO. Thanks to the high clock speed, though, it can be expected that it will be only slightly behind, if at all. Compared against its predecessor, the GeForce4 Ti4200, the FX 5600 Ultra scores points only with FSAA and anisotropic filtering. In the standard tests, the card is clearly behind most of the time. For those who own a GeForce4 Ti4200 and are considering an upgrade, you're only going to get limited satisfaction out of the FX 5600 Ultra. A 9500 PRO is the better choice here, because of its gains in FSAA/ anisotropic as well as raw performance. You can also expect the same to be true for the 9600 PRO. Another surprising aspect is the FX 5600U's poor vertex shader performance in 3D Mark 2001/ 2003. The results of the pixel shader test are not terribly impressive, either.
The gap between the FX 5600 Ultra and the FX 5800 is a big one - too big, actually. Here, NVIDIA was apparently overzealous in wielding the red pen!
By contrast, we were more pleased by the FX 5200 Ultra. With respect to performance, the card can hold its own quite well against the Radeon 9000 PRO/ 9200. As a DirectX 9 card, it is already superior to the others on paper, and it also offers multi-sampling FSAA and relatively fast anisotropic filtering. Because of ATi's driver problems, it wasn't possible to make direct comparisons with the 9000/ 9200 series, which still use the slow SuperSampling, but from past experiences with the 9000/ 8500 FSAA test, it can be concluded that the FX 5200 Ultra should be the better performer here. The gains compared to the GeForce4 MX440-8x are quite clear as well. In the entry-level segment, the FX 5200 Ultra is therefore a good choice. It's not certain, however, as to whether the moderate performance of the card really allows the DirectX 9 features in games to be used. Tests with modern games such as Splinter Cell or Aquanox 2 (internal tests with DX 9 Beta Patch) show that the card is actually too slow for this.
We'll only be able to draw a definite conclusion about the GeForceFX 5600 when ATi launches the new Radeon 9600 series. But compared to the Radeon 9500 PRO (priced about the same, at $199), NVIDIA's mainstream flagship comes up short. In addition, there's the FSAA/ anisotropic image quality of the ATi R3x0 chip , which turned out to be significantly better in our previous test. The only thing that can save NVIDIA from more serious damage is the poor delivery situation that is typical with ATi.
