Inno3D Tornado GeForce3 in the test: GeForce 3 against Geforce2 Ultra
- 1 Introduction
- Scope of delivery
- 2 Technical data
- 3 New features
- nfiniteFX Engine
- Vertex Shader
- 4 Pixel Shader
- Light Speed Memory Architecture
- 5 Test configuration
- 6 3DMark2000
- 7 3DMark2001
- 8 Unreal Tournament (utbench)
- 9 Unreal Tournament (benchmark)
- 10 Quake III Arena
- 11 Vulpine GLMark
- 12 GL Excess
- 13 FSAA benchmarks
- 14 2x FSAA duel
- 15 4x FSAA duel
- 16 Overclocking
- Overclocking Benchmarks
- Overclocking Conclusion
- 17 Conclusion
With the Pixel Shader, the finest details can be rendered in real time. This means that completely new materials can be represented, which makes organic materials such as skin look much more realistic. The best example to illustrate the pixel shaders (in this case the structured reflections) is the X-Isle demo. It shows a landscape from the primeval times of our earth, in which some dinosaurs are up to mischief. To demonstrate the visual difference, the same dinosaur was recorded without the Gerforce3 features. To the right of it again an excerpt from a lake from the same demo, on the right with Geforce3 features, on the left without.
Environmental bump mapping will be one of the most common effects used in future gameswill. Matrox introduced this technology almost two years ago with the G400, albeit not with the same performance as the Geforce3. ATI also implemented this technology in the Radeon.
Light Speed Memory Architecture
In order to be able to counter the demand for ever increasing memory bandwidth and to be able to better use the available memory bandwidth, The Geforce3 brings with it a whole host of new technologies called Light Speed Memory Architecture.
- Crossbar Memory Architecture With a 128-bit DDR memory interface, the memory is used Controller data in 256 bit 'pieces' (DDR 2x128 bit!). In contrast to its predecessors, the Geforce3 chip allows four individual 64-bit accesses. This increases the efficiency of the available bandwidth enormously.
- Lossless Z-Compression The Z-value of a pixel stands for its depth information in the current image. The graphics chip has to read out its Z-value for each pixel and, if necessary, also rewrite it, which of course puts enormous strain on the memory bandwidth. In order to save memory bandwidth there, nVidia now uses lossless 4: 1 data compression.
- Z-occlusion culling method The Z-occlusion culling method is used to reduce the available memory bandwidth also spared. If the Geforce3 chip assumes that a pixel is invisible, it will simply no longer be rendered. However, this should not be confused with the scene-based rendering of the Kyro, since the Geforce3 checks during depth rendering whether the pixel has already been covered, but pixels that would later be painted over again are drawn anyway.
- High Order Surfaces 3D graphics have always been represented by many triangles. However, this has the disadvantage that a lot of triangles have to be drawn, especially for curves, bulges or curves, so that theSurface looks smooth. Professional 3D programs already use so-called splines, according to which the curve is created with a formula and control points. A game can now transmit the data of the spline to the Geforce3, since the Geforce3 now also supports this in the hardware.
- High Resolution Antialiasing (HRAA) With High Resolution Antialiasing, so-called Step effects on objects are eliminated, which basically means that virtual pixels are placed in the steps. The Geforce3 no longer does this through supersampling, but rather through multisampling. The difference in performance is simply that 1 texture value is no longer used per pixel (e.g. made 4 texture values for FSAA x4). The Geforce3 just takes one texture value for all subpixels (with FSAA x4 only one). The effect is limited to the anti-aliasing, a texture flicker, which was also reduced with supersampling, remains untouched. Quincunx is also new, a new sampling mode that promises the quality of the x4 mode for the performance of the x2 mode, which mixes 5 subpixels for the final image, but still only calculates 2 subpixels of its own per pixel and therefore technically belongs to the FSAA x2 series belongs. In fact, it is a bit slower than the x2 mode, but the quality is really close, if not better, than the x4 mode, but the drop in performance is still below the x2 mode of the GeForce2 Ultra. According to nVidia, the HRAA of the Geforce3 should be compared to the Geforce2 Ultra. be up to four times as fast, which we will check later in the benchmarks. The following image section from Strikeforce demonstrates the qualities of the individual antialiasing modes of the Geforce3 very well.
On the next page: Test configuration