CONTENTS

1. Introduction
2. R420/NV40 features
3. Configurations of testbeds, test tools
4. Filtering quality: FarCry
5. Filtering quality: Need For Speed: Underground
6. Filtering quality: Pirates of the Carribean
7. Filtering quality: Unreal II
8. Filtering quality: Unreal Tournament 2004
9. Filtering quality: Serious Sam: The Second Encounter
10. Filtering quality: Painkiller
11. Conclusions
    
    
    
    

Conclusions

Our previous material dealt with the differences between X800 and GeForce 6800 filtering realisations in a RightMark synthetic test. As you remember, the picture was quite ambiguous and in general, 6800 trilinear filtering optimisation brought about a somewhat lesser quality fall than X800, although a parity could also be recognised. But the "mistake" in the control panel (mentioned by ATI officials six months ago) has never been corrected. Thus, when anisotropy is forced in applications that can control this function, trilinear filtering vanishes into thin air.

Concerning the games, they have shown that there are but slight differences between various realisations of the mentioned filterings.

Certainly, we have examined only a small part of modern games, and our June 3DiGest will give a more detailed account of this aspect. But still, we can say that there is no visible difference in graphic quality if bilinear filtering or bilinear plus trilinear ones are active. What is really different is that borders between MIP levels are not so radically smoothed any longer, and you can only notice the smoothing on small marginal areas. Sure, you can easily notice it in the games featuring a lot of single-tone textures that cover large areas. Optimisations can really spoil your perception of the picture there.

But you'll hardly find such things in contemporary games. Either the textures are motley and selected so that borders between MIP levels are virtually unnoticed, or the relief is so complex that you don't see MIP mapping. There are but few places where enabling anisotropy reveals problems with trilinear filtering.

We leave it up to our readers to decide what is better, but I personally consider the approach used in GeForce 6800 more correct, and simplified trilinear filtering is less visible there. And besides, if the user forces anisotropy in a game that can manage the function itself, then its realisation will be correct and correspond to the norms on GeForce 6800, while it will disappear whatsoever on X800. FarCry can serve as a perfect example for this. The game only allows four degrees of anisotropy, which is not enough for users of powerful cards. Thus, any owner of an expensive accelerator will set anisotropic filtering in the drivers. As a result, GeForce 6800 will work correctly while RADEON X800 will turn off trilinear filtering. Surely, ATI fans will be quick to remind me of certain problems related to NV3x/4x's shader calculation precision in this game. Well, that's really a drawback, but patch 1.2 is to appear soon and we'll see if it can correct it.

Summing up, I would like to say the following thing. Despite multiple accusations that ATI or NVIDIA deceives users, the main problems lie not in the realisation of trilinear filtering or even anisotropy. They are driver/game bugs that result in real artefacts and not in some trifles found by a meticulous admirer of the rival company. Open our 3DiGest section that deals with quality in games and you'll see that there are claims about many other things which bad programmers are really to blame for (including game developers as it's their mistakes that cause almost a lion's share of all problems).

Therefore, no matter how indignant fans of "pure graphics" are (in fact, they are usually fans of either ATI or NVIDIA), an overwhelming majority of users will see NO DIFFERENCE in most games due to all these optimisations.

It is hardly any good lighting MIP levels just to point at the cheater. Only a tiny fraction of users will ever do such thing. Sometimes, it only makes the developers use such optimisations, often without users knowing about it.

So, is it good or bad? I think everyone should be able to choose. Even if a user never sees the difference and never enables optimisation, he still deserves to have a choice. If a washing machine lowers the frequency on the Q.T. and underwrings out the clothes, most householders will hardly notice it, but inevitably, there will be some people who will make a huge row out of it. And the manufacturing company will have to apologise for this unintended effect (as was the case with ATI). But if that washing machine had a mode regulator, each could choose the frequency himself and would be aware of his actions. Then there would be no claims, and NVIDIA would have no reason to issue humiliating presentations concerning ATI products. In the end, the Californian company turned out to be fairer, although it used to be the main focus of reproach for optimisations.

You can see more detailed comparisons of various videocards in our 3Digests.

[ Previous part (8) ]

Article navigation: