There is one important point not brought up:
Texture sizes should be square and power of 2.
Especially older cards like GF4 and Radeon 9250 and previous, while supporting non square power of 2, gain a serious amount of performance if you feed them with square textures! (30% and more)
B3D will resize the texture to be power of 2, but not resize it to be square (on GF 6600 this can even lead to crashes, on some 6800 as well, depending on driver versions)
Another thing that speeds up is using DDS instead of regular textures where possible. They are compressed even in VRAM and makes them faster when transfering into VRAM for rendering.
Texture sizes should be square and power of 2.
Especially older cards like GF4 and Radeon 9250 and previous, while supporting non square power of 2, gain a serious amount of performance if you feed them with square textures! (30% and more)
B3D will resize the texture to be power of 2, but not resize it to be square (on GF 6600 this can even lead to crashes, on some 6800 as well, depending on driver versions)
Another thing that speeds up is using DDS instead of regular textures where possible. They are compressed even in VRAM and makes them faster when transfering into VRAM for rendering.