Why should texture size be less than power of 2?

Question

Hi!
I just read in the packagerforiphone_devguide.pdf
Make bitmaps in sizes that are close to, but less than, 2^n by 2^m bits. The dimensions do not have to be power of 2,
but they should be close to a power of 2, without being larger. For example, a 31-by-15–pixel image renders faster
than a 33-by-17–pixel image. (31 and 15 are just less than powers of 2: 32 and 16.) Such images also use memory
more efficiently.
can someone explain me this? As Gamedeveloper i'm used to have power of 2 textures. why should they be smaller than power of 2 instead of exactly power of 2?

Answer 1

I have read that too, and it doesn't make sense. The article would have been a lot clearer if they had used 32x16 as the second example, instead of 33x17. Like you, I would almost expect advice that said to not exceed a power of 2, but it does't seem right to say that the exact power of 2 isn't recommended either.
Hey, I just found an earlier passage that pretty well states that we're right, on page 34:
"The GPU uses memory allocations that are powers of 2 for each dimension of the bitmap image. For example, the GPU can reserve memory in 512 x 1024 or 8 x 32 sizes. So a 9 x 15-pixel image takes up the same amount of memory as a 16 x 16-pixel image. For cached display objects, you may want to use dimensions that are close to powers of 2 (but not more) in each direction. For example, it is more efficient to use a 32 x 16-pixel display object than a 33 x 17-pixel display object."
So there you go, a 32x16 bitmap does use less memory than a 33x17 bitmap.
I think, or hope, that the later part was just making the point that you don't have to use exact powers of 2, but in doing so you want to not go over the power of 2. Hence using examples of 31x15 and 33x17. They didn't use 32x16 as an example because those are powers of 2, which wasn't what was being illustrated.