On OSX10.4/iOS5 and onwards you can optimize your texture uploads and downloads using CVOpenGL[ES]TextureCaches.
Instead of uploading textures with glTexImage2D and reading from the frame buffer with glReadPixels, you use a CVOpenGL[ES]TextureCache to translate your texture/FBO operations into the language of CoreVideo CVPixelBuffers.
This works perfectly well with byte (and probably short) sized integer formats, but, apart from a fancy YUV pixel format, floats are decidedly under-represented in Core Video's Pixel Formats.
Does this mean I can't use GL_FLOAT textures and render targets with CVOpenGL[ES]TextureCaches?
