The GeForce 4 Technology Brief
There's a lot of info surrounding the GeForce 4 technology, and although there's too much to mention on one page, we'll explain what we can. For additional info, feel free to read this . For now, here is what you have to know. The GeForce 4 we're looking at today is the Titanium 4600. The other members of the Titanium family are the Ti4400, and Ti4200. Both are slower parts, clocked 25MHz, and 50MHz slower respectively, when compared to the top of the line Ti4600. All three cards share the same feature set, and are based on the NV25 technology. The GeForce 4 MX is based on NV17 technology though, and if you are still uncertain, you can read our GeForce 4 MX debrief here, as well as our MX 440 review, and also our MX 420 review. So, what are the main features of the NV25? We'll go through each of them...
nfiniteFX II Engine
Introduced with the GeForce 3, nVidia has upgraded their nfiniteFX technology in several ways. To begin with, everything is just a lot faster than before. It's not all just speed though. nVidia have added a second vertex shader which will allow for more vertices per second (136 Million of them to be exact). This feature, along with more advanced pixel shaders, will allow for more realistic, and faster animation. As with the GeForce 3, all of this is programmable, meaning developers aren't restricted to code around hard coded rules in a GPU.
To better explain, a vertex is the corner where two edges of a triangle meet. Because polygons are made up of triangles, being able to change their values will create a more realistic scene. Pixel shaders is used to create textures and surfaces. Rather than placing a bitmap on an object, pixal shaders can modify values on a per pixel basis.
A good example of this technology at work is their Wolfman demo. Drawing fur is a lot more complicated than a shiny surface such as plastic, and on any lesser technology, this would grind a system to a halt. Compared to a GeForce 3, it's pixel shader performance is at least 50% faster than before.
The core of the NV25 is fast, but the bottleneck in video card performance has always been the memory. The Ti4600 clocks it's ram at 325MHz (650MHz DDR), which is an improvement of 75MHz over the GF3 Ti500. With that, improvements have been made to it's Lightspeed Memory Architecture (II), Lossless Z-Compression, and Occlusion Culling.
Just as with the GeForce 3, the Crossbar Memory Controller is still present, where four separate controllers act as traffic cops for the bandwidth highway. The controller avoids bombarding the AGP bus with texture information, and makes it much more efficient, allowing for a maximum bandwidth of 10.4GB/Sec. In case you're wondering, that's a lot. If you're wondering what needs that much bandwidth, it's two things. Antialiasing, which we'll get to soon, and high resolution gaming.
Lossless Z-Compression compresses Z data, and delivers a 4:1 benefit. This feature saves memory bandwidth, and will allow for faster memory performance.
Occlusion Culling, or hidden surface removal, was something introduced with the Kyro cards, used in the GeForce 3 series, and improved with the GeForce 4. Simply put, previous video cards would render an entire scene, even when an object isn't displayed. This eats up bandwidth, and slows things down. Occlusion Culling will allow for better performance because it won't render something you cannot see.
Accuview Antialiasing (AA)
Antialiasing is best described as "removing the jaggies". 3dfx pimped their Full Screen Antialiasing with the Voodoo 5, and every major video card player offers it now as a feature. nVidia introduced their high-resolution antialiasing (HRAA) last year, and although Accuview does the same thing (getting rid of jaggies), it does it differently.
Unlike before, where specific points of a pixel(s) were selected for sampling, what the new technique Accuview does is pull random spots of a pixel. This method is not as rigid, or as mathematically correct as before, and the end result is a better looking image. Because everything is faster now, AA gaming is more playable, and with the addition of dedicated hardware multisampling on-die, the AA architecture is just plain meaner.
All the standard AA methods are available, be it the supersampled 2x, 4x or the multisampled Quincunx, but nVidia has introduced a new multisampling technology called 4XS, which is a Direct3D feature only. Because I'm an OpenGL snob, I was unable to test 4XS extensively, other than running our standard 3D Mark 2001 SE benchmarks.
nView Display Technology
Previously introduced with the GeForce 2 MX, nView makes it's appearance on the entire GeForce 4 product line. Granted, the manufacturer has to package it, but for graphic professionals who are closet gamers, you now have this option. You can work in your main Photoshop canvas on one display, and put all your palettes on another, increasing viewable desk real estate. VGA + VGA, VGA+ DVI, and DVI + DVI are all supported by Integrated Dual 350MHz DACs, and Integrated Dual-Channel TMDS Transmitters for DVI. nView also supports TV-out at 1024x768 resolution, thanks to the Phillips Integrated TV Encoder.
You can read our look at nView here.
Prev Page - The card
Next Page - Overclocking and Benchmark Setup