Revolution over Evolution
So, what's so revolutionary about the latest from NVIDIA ?. Let's take it step by step.
The NVIDIA GeForce FX range of video cards will centre around what NVIDIA dub their 'CineFX' engine. What's this ?, and how does it enhance the overall graphical function. Let's have a look at the constituent parts that can be covered by the umbrella term that is Cine FX.
It seems as if the goal of the latest graphics cards appears to be to close the gap between cinema-quality and PC-quality graphics. Achieving such a lofty aim is inherently difficult. You need support on both a hardware and software level, as well as a truly programmable GPU. The use of increasingly complex shading will help to narrow the gap between present and cinema-quality presentation. Add to this a massive bandwidth requirement and the need for an efficient architecture, then you'll see why it hasn't been easy to implement till now.
In terms of API support, the GeForce FX goes above the requirements laid down by Microsoft's DirectX 9.0 API. The following chart should explain things a little better.
As you can no doubt see, the GeForce FX comfortably exceeds the standards laid down by the API. The idea is to have a fully flexible and programmable GPU, thereby allowing one to create truly cinematic-style graphics. Complementing this increase in shading ability is NVIDIA's in-house CG programming language. Previously, writing complex shading code required cumbersome assembly code. The far more subtle CG language, as pushed by NVIDIA, should make this task a great deal easier. It's not just the fact that the GeForce FX has more instructions and options for shading than all that have come before, it's the way it applies them.
Traditional shading is done on a per-object basis. What that essentially means is that you have to split the given object into constituent parts, shade them, and draw them separately. With the NV25-class of GPU that means a shader for each part. With a R9700 and DX9, it's improved to the use of a single shader for all the parts, but you still have to split and segment your model into its constituent parts and draw separately. With the FX GPU, shading and branching is not done on an object basis, but rather a vertex basis. Additionally, one shader can be written that encompasses all the skinning methods and operations. Because shading is done on a per-vertex basis, there's no need to split a model up into different parts, making for faster overall shading.
To complement the increase in shading available, and to make graphics far more lifelike, NVIDIA will support both 16 and 32-bit floating-point colour formats. Studio-quality colour requires the use of 128-bit colour, this can be achieved by using the 32-bit FP (32-bits for each RGB component) on offer. Just imagine the quality of various effects such as fog if 128-bit precision is used. This, however, isn't limited to solely the FX. ATi's R9700 Pro features this level of precision, too. Precision is everything if you're striving for cinema-quality graphics.
