Most graphics cards work faster when the image size is a power of 2 (128, 256, 512 etc.)

Imagine the low level code require for accessing images that were exactly 497 x 363!

Even though that may be all you need out of your texture, it's faster for the processor if it was working with 512 x 512 blocks with some wasted space in them.

Power of 2 multiplication or division is a simple shift left or shift right (respectively):

256 << 1 == 512
512 >> 1 == 256

Instead of having to call a (typically CPU costly) divide or multiply opcode, you can get away with a simple bit shift which typically only takes a single CPU cycle.


The choice between using say 256 x 256 or 1024 x 1024 is arbitrary and usually depends on how close the camera may get to the textured surface (we've all seen the 3D games where the walls get blocky when you move too close to them...)

OR

If the same texture map is being used by more than one object - you'll want to have more "space" to work with using a larger image map - it all boils down to image quality vs available texture memory - the larger the image, the better it will look...

-Will