I've already covered the features of the 9800 XT in my 9800 Pro review. If you want to read it again, read my 9800 Pro review on Motherboards.org click this link. The major differences have already been covered in the Card section (clock speed, cooling, overclocking) and don't need repeating here. The newly released Catalyst 3.8 drivers however, deserves special mention since they added quite a few new features.
Some mention should be made about the pixel shaders on the 9800XT. Much has been made about the "optimizations/cheating" by NVIDIA in current benchmarks. Part of the reason for the need to do better in benchmarks was that ATI has a better overall architecture. ATI says that their architecture does 40 pixel shader operations in a second, compared to the 12 of NVIDIA's NV30 architecture. The R3xx architecture can output 5 pixel shader instructions in a clock per pipeline. NVIDIA's NV30 architecture on the other hand is sort of a hybrid 8/4 pipeline design. With the ability to do 3 pixel shader instructions in a clock per pipeline that gives a total of 12 instructions a clock.
The 9600XT feature set is also well described in my 9800 Pro review. ATI added low-k dielectrics to the process of the chip which allowed power consumption levels to go way down. Also notable was the increase in clock speed of the 9600XT to 500/300 core/memory from the 9600 Pro's 400/300. Unfortunately, I don't have a 9600 Pro to test against the 9600XT, so the performance is going to look a bit strange.
