The anisotropy quality of the RADEON 9700 is excellent at the angles of 0, 45 and 90 degrees and it is higher than that of the GeForce FX 5800, but it also falls down markedly on the planes parallel to the angles of 30 and 60. The NV30'a quality is more or less the same everywhere. In Direct3D. But what do we have in OpenGL? 

As you can see, the anisotropy in the NV30 on far distances is like being cut-off (although it is more similar to disappearance of trilinear filtering, it still exists). I think that it is either because of the drivers, or the developers have simplified anisotropy in the OpenGL. No more complains. There are no screenshots at different angles as there are no problems here (if you suspect that NVIDIA might use the anisotropy from RADEON 8500, do not worry - it will hardly ever happen). 

ANTI-ALIASING (AA)

At the beginning we studied the aspects of the AA operation, then estimated performance in the previous section and now let's deal with the quality. 

1. Almost no quality difference between 4x and 6xS/8xS. So, why to pay more for the same? 
2. The R300 has higher AA4x quality than the NV30/NV25. 

3D quality in general

In general, I have no complains. I just can show you the screenshots from NFS:HP2 obtained with the NV30: 

Only in case of AA 8xS there are artifacts like these: 

I think it's the driver's fault again. 

Conclusion

At last, we managed to test this so long-awaited NVIDIA's solution which is the fastest and most flexible accelerator for today! However, the cards are not available on the market, and it is possible that the GeForce FX 5800 Ultra will never come onto the scene (maybe only GeForce FX 5800 working at lower frequencies and the slower RADEON 9700 PRO). 

But the NV30 is a real step toward a very flexible programmable graphics processor. The GeForce FX has its disadvantages which were mentioned above. But they do not lay deep and I suggest that in the next chip versions of the NV3X family (first of all, NV35) the errors of the NV30 will be done away with. I have got information that the NV35 may arrive already in May. Besides, we expecting the junior brothers - NV31/34 which will also include all shader capabilities of the NV3X architecture. 

1. It's the first time we are examining the card which may not arrive into the retail market but which serves an an architectural base for a whole family of products. The already tested execution units based on the .13 technology will let quickly make and launch a whole family of new chips for various market niches. 
2. The weakest point of the GeForce FX 5800 is the 128bit memory  bus because even such fast memory working at 500 (1000) MHz can't provide the required throughput for the 500MHz 8-pipeline chip. It had a very bad effect on the AA in spite of the frame buffer compression in MSAA modes. Such compression is the de-facto standard for modern processors, and even with this compression the throughput of the 128bit memory bus is not enough for storing 4000 shaded pixels per second. As a result, the NV30 has a lower speed in the AA mode than the R300 (which has both compression and the 256bit memory bus). Besides, the MSAA algorithm's quality is much higher in case of the R300. 
3. We can't estimate whether the GeForce FX 5800 Ultra is able to cover the expenses with sales because the final retail prices are unknown. 
4. Another downside is a noisy cooler which can frighten users by unexpected starts or pauses (when the rotation speed gets lower). Because of such design a user has to sacrifice the first PCI slot. But this downside can be dealt with, and there are already cards with cooling systems free of shortcomings of the reference card. 
5. So, despite the above weak points, this is the fastest 3D accelerator for the beginning of 2003. In our 3Digest we will see how the NV30 works on different driver versions and check quality and stability in games. 
6. In this review we didn't touch on operation of the card with TV and DVD playback, but in the nearest review on a production NV30 based card we will make up for it. 

Article navigation: