When software is written to take advantage of it, hardware T&L offloads the computation-intensive T&L tasks from the CPU, freeing the CPU for other tasks. Also, NVIDIA's dedicated T&L engine is faster than current CPUs for T&L tasks, so more complex scenes can be drawn with it. There is a complex scene using hardware T&L pictured below with textures and in wire frame mode.
A complex scene rendered with textures
The same complex scene without textures
The MX sports two rendering pipelines each capable of processing one pixel with two textures per clock. Two pixels per clock at 175MHz gives the MX a 350Mpixel fill rate. And two textures per pixel at 350Mpixels gives the MX a 700Mtexel fill rate. In comparison, the GTS sports four rendering pipelines, each capable of processing one pixel with two textures per clock. Four pixels per clock with two textures per pixel at 200MHz gives the GTS a theoretical 800Mpixel fill rate and a 1600Mtexel fill rate. At this point we should also bring up the 256, which has 480Mpixel and Mtexel fill rates. So in the area of theoretical power, the MX lies somewhere closer to the 256 and well under the GTS.
