I modelli a confronto:
L'architettura:
Lightspeed Memory Architecture II
Lightspeed Memory Architecture II (LMA II) is a continuation of Lightspeed Memory Architecture implemented in GeForce3.
LMA II includes a 128bit memory controller divided into 4 32bit ones, like the one used in NVIDIA GeForce3, and some other technologies aimed at increasing the efficiency of the graphics memory. They are:
Z-Occlusion Culling. This method was first introduced in NVIDIA GeForce3. It allows determining if the pixel is visible by its Z-coordinate, before the texturing is done. And if the pixel is invisible in the scene then it will not be processed thus saving time on addressing the frame buffer, Z-buffer and textures.
Lossless Z Compression. This technology is also implemented in both: NVIDIA GeForce3 and NVIDIA GeForce4. It allows compressing the Z-buffer as 4:1 this way saving quite a lot on data transfer and storage in the Z-buffer.
Fast Z Clear. This function allows clearing Z-buffer and filling it with the initial Z value before building every new frame. NVIDIA claims that this function was for the first time implemented in GeForce4.
Memory Auto Pre-Charge. This technology prepares the graphics memory device proactively to pre-charge areas of the memory that are not in use now but are likely to be used in the very near future. NVIDIA claims that it helps avoid forced idling of the GPU lasting up to 10 clocks, when the GPU is actually waiting for the graphics memory to get ready for further action. The wait state in case of Auto Pre-Charge drops down to 2-3 clocks, as NVIDIA engineers claim.
Quad Cache. NVIDIA GeForce4 features 4 independent buffers (caches). Individual caches store primitive, vertex, texture, and pixel information. These caches are individually optimized for the specific information they deal with.
Accuview and Full-Scene Anti-Aliasing 4xS:
Different versions of the full-scene anti-aliasing involving multisampling, 2x, Quincunx, 4x and now one more, 4xS technique, together with textures anisotropic filtering are now called with one single word - Accuview.
However, there is only one thing really new here: 4xS Full-Scene Anti-Aliasing technique.
In fact 4xS technique, which was actually working by NVIDIA GeForce3 as well, is a combination of 2x multisampling and 1x2 supersampling. The final pixel color in case of 4xS is obtained as an average of the two subpixels located one below another, each of which is formed by 2x multisampling. To understand it better, imagine a "2:2" domino piece standing vertically.
4xS technique smoothens the "steps" on the horizontal and slightly inclined lines. We checked it with the help of Serious Sam: The Second Encounter, where we compared the quality provided by 2x, Quincunx, 4x and 4xS techniques from NVIDIA. For a more illustrative example, we also added the results obtained for SMOOTHVISION 6x Quality from ATI:
No Anti-Aliasing
Accuview 2x
Accuview Quincunx
Accuview 4x
Accuview 4xS
ATI Smoothvision 6x Quality
Qualità d'immagine:
GeForce4 TI 4600
GeForce3 TI500
Ati Radeon 8500
Benchmark