Using gluBuild2DMipmaps and an appropriate minification filter can help too, and still let you have the full-size texture when viewed close-up.

- Alex Diener

That didnt seems to work before, im probably doing it wrong but can you do it with textures that have alpha channels?

You can, but you should know about texture halos and premultiplication first. A skim through this thread should explain it: http://www.idevgames.com/forum/showthread.php?t=7560

When you tried it before, what parameter did you use for GL_TEXTURE_MIN_FILTER?

- Alex Diener

GL_LINEAR for MIN and MAX
Halos arent a problem but premultiplication sounds like a good idea, will that speed things up at all.
(does it count as premultiplying if Im loading pngs with alpha channels?)

Premultiplication won't make a difference in speed. It's mostly just useful for eliminating halos. When you load a PNG, unless it's already premultiplied or you do it yourself, it won't be. If you were using premultiplied alpha, you'd want to set your blend func to GL_ONE, GL_ONE_MINUS_SRC_ALPHA.

When you use GL_LINEAR as your minification filter, it'll take the full-size texture and scale it, ignoring any mipmaps that are available. To get it to use mipmaps, you'll want to use one of GL_LINEAR_MIPMAP_NEAREST, GL_LINEAR_MIPMAP_LINEAR, GL_NEAREST_MIPMAP_LINEAR, or GL_NEAREST_MIPMAP_NEAREST. I find that GL_LINEAR_MIPMAP_NEAREST generally gives me the best quality | speed ratio, but you may want GL_LINEAR_MIPMAP_LINEAR for the nicest possible results. Any one of those, however, is likely to be significantly faster than GL_LINEAR.

- Alex Diener

Ahhh, i see thats very clever.
Im confused about Nearest though, I dont want to use a 128x128 texture for a 150x150 quad when I could use the 256x256 texture.
What I mean is I want to always use a texture of a higher resolution than the quads, how could I do that?

I'm not entirely sure how MIPMAP_NEAREST works myself, but looking at the man page for glTexParameter, it seems that would indeed be the behavior it would give you. It doesn't look like there's a way to have it always use the next largest mipmap level, but GL_LINEAR_MIPMAP_LINEAR should work OK for you. Give it a try and see if it gives you the intended effect.

- Alex Diener

You can bias the LOD parameter to force a larger/smaller mipmap level to be chosen, with the EXT_texture_lod_bias extension.

Ill look that up, thanks