Asus V8440 TD and V8460 ultra TD in the test: Two 'Ti' tans among themselves
- 1 Introduction
- 2 The cards
- 3 Scope of delivery
- Drivers and tools
- 4 Technical details
- 6 Test system
- 7 Synthetic benchmarks
- 10 Game benchmarks
- 15 FSAA Performance
- 18 Anisotropic Filter
- 20 FSAA and AF combined
- Image quality
- 21 Conclusion
Now we come to the subject of full screen anti aliasing. Since the introduction of the GeForce3, nVidia chips have not only mastered the classic supersampling process, which internally calculates the image completely in a higher resolution and then itscaled down and output to the RAMDAC via the framebuffer. In terms of quality, the process is very high quality, but unfortunately requires a lot of power, which is not surprising, the process causes even more effort than actually playing in a higher resolution. For performance reasons, GeForce3 and GeForce4 (MX and Ti) use a process called multisampling, which only works at the edges of the objects in a scene. The advantage of this is that the annoying podium artefacts are now smoothed out with significantly increased performance. The disadvantage is that the textures remain completely untreated.
From left to right you can see the same section from an image without FSAA, with 4x multisampling and with 4x supersampling. The described effect can be seen particularly well on the texture of the roof, but the soil texture below the black bar (a drawbar) is also clearly sharpened from the supersampling process.
The pixel flickering and the increased texture sharpness with classic supersampling have to be achieved with GeForce3/4 with other means. One of these means is anisotropic filtering, the addition of which unfortunately costs performance again and is only available in one level with the GeForce4MX, while the GeForce3 and GeForce4 Ti offer four levels.
The chips of the GeForce4 series also have the advantage of having one with 2xMSAA and with Quincunx-AASave write access to the framebuffer, but according to nVidia not, as previously claimed, mix the image together in the RAMDAC. Unfortunately, the anti-aliasing effect cannot be seen on screenshots that were created using the 'Print' button and thus copied the content of the frame buffer. This is just a preliminary remark, before we come to the pure performance show.
3DMark 2001 SE + FSAA
For the benchmarks for the FSAA, we selected a few very special candidates and measured them. Due to the extremely large number of measured values required, the originally created diagrams were unfortunately hardly usable due to their distribution over several screen pages. That's why we switched to the tabular representation.
First the 3DMark2001 SE, in which the GeForce4 MX440 has to do without the Nature test points as always because it lacks the pixel shader and the vertex shader functionality that is not relevant here.
The gradations between the Geforce4 models are almost linear, even if they are Top model can clearly set itself apart. It is interesting to see that the Asus V8460 ultra 4xS-FSAA is more or less 'for free' if you take the value of the MX440 without FSAA, which is a good quarter as expensive. The higher resolution can also be used in this resolution without FSAAChiptakt make the decisive difference in favor of the Geforce4 Ti4200 (250/222MHz), with activated FSAA, however, the sheet changes drastically and even in the next higher resolution memory bandwidth cannot be replaced by anything.
You can already see the significantly higher demands that the 1152 resolution both on the Speic Both the interface and bandwidth and, in the case of the GeForce3 and GeForce4 TI with pixel shaders, depend on the chip clock, since the difference to the shaderless MX440, which only loses 500 points without FSAA and less than 400 points with 4xS-FSAA, is already decreasing .
Here at the latest it becomes clear that the GeForce4 Ti4200 with 128MB, i.e. the 250/222MHz clock rate in high FSAA modes (4x/4xS) is not really significantfaster than a GeForce3 let alone a GeForce3 Ti500. The GeForce4 MX440 was removed from the competition at this point because, even with the best will in the world, no acceptable performance could be achieved in the 3DMark2001 SE due to the high resolution and FSAA, especially in the higher quality 4x FSAA modes.
4xFSAA in 1600x1200x32Bit remain a domain of the cards equipped with 128MB due to the pure frame buffer size. The 64MB equipped cards refused to work here. Throughout the bank up to the resolution of 1280 it could be seen that the promise given by nVidia that Quincunx-FSAA would now cost no additional service compared to 2xFSAA had been kept. Only in the King's resolution does the available fill rate increasingly appear as a limiting element and cause an additional loss of performance.
On the next page: Quake III Arena + FSAA