If you search around on the web you can probably find many documents describing how video card processors work in excruciating detail.
Simply put, a 'pipeline' is a path for data to take through a processor. All graphics cards today use multiple pipelines in tandem, allowing them to work on several pieces of data at once. This is cheaper (and easier, especially in terms of drivers) than having multiple GPUs on a single card, but not quite as fast as doing that.
The R3XX-series cards from ATI use either 4 pipelines with one texturing unit apiece (4x1, found in the R9000-R9600), or 8 pipelines with 1 texturing unit apiece (8x1, in the R9700 and R9800).
NVIDIA's high-end cards use a design with 4 pipelines with two texturing units apiece (4x2). In theory this should be as fast or faster when dealing with multitexturing situations (where you are computing things involving more than one texture being applied to a surface simultaneously). However, in single-texturing situations, it's not going to be nearly as fast.
That's a VERY dumbed-down description of how this all works; try searching for older threads on this topic if you're interested in more gory details.
