A second technique implemented in the Lightspeed Architecture is a lossless Z compression algorithm - a feature straight out of our RADEON Guide. The algorithm compresses Z-buffer traffic (the data that defines where an object sits on the Z axis in 3D space) by a factor of four and is compressed and decompressed on the fly without a performance penalty. Given that the algorithm is lossless, none of the Z precision is sacrificed.

Finally, NVIDIA has added a feature called Z-Occlusion Culling that attempts to determine whether or not a pixel is going to be displayed or if it will be covered by another object (like ATI's Hierarchical Z). If it will indeed be covered, the GeForce3 will not render it, will not make those calls to the Z buffer, and the bandwidth which would have been used can be re-purposed. On average, 3D applications have a depth complexity of two, meaning for every visible pixel, two have to be rendered to achieve the result. If implemented properly, Z-Occlusion Culling has the potential to save some serious bandwidth.

When the GeForce2 GTS was released, "pixel shaders" were the buzz. Now, NVIDIA is parlaying a new technology aptly named "vertex shaders." The underlying purpose of this programmable vertex pipeline is to deliver a virtually infinite number of real-time visual effects. In it's most basic form, the vertex shader is a graphics processing function that adds special effects to objects in a 3D graphics scene. A programmable vertex shader, like that used in the GeForce3, allows developers a new level of flexibility. Vertex data, which can consist of x, y, and z coordinates in addition to color, lighting, and texture instructions is fed into vertex shader - a box, if you will, containing a mathematical function. The function manipulates the vertex data (it does not delete or create any data) and the changed vertex data emerges on the other side with different coordinates, different transparency or a different color. Not every vertex that enters the vertex shader is necessarily changed - the function can specify only those vertices with certain properties. Once the function runs for a specific special effect, it can also be unloaded from the vertex shader to make room for a second effect, which can be applied to the same scene.

As you can see from the above image, the vertex shader runs in parallel with the hardwired T&L unit of the GeForce3. If the vertex shader is running, the T&L unit is idle. This isn't an issue, however, because the output of the vertex shader is a fully transformed and lit vertex. DirectX 7 applications that take advantage of static T&L will run through the hardwired T&L circuitry, while DirectX 8+ applications programmed to utilize the vertex shading circuitry will not utilize the hardwired T&L. This ensures compatibility between both old and new software.

Given that the functions supported by the vertex shader are programmable, the number of available effects are limited only by the creativity of the developers who choose to program for the GPU. To give the developers a head start, NVIDIA has written a special application to manage the effects library and included nearly 100 ready-made effects. We were given the opportunity to witness some of these effects, with some pretty impressive results.