I have HD 5670, and in my test there is 250 textures and 500 rectangles that I render 10 times in a frame. Without using sorting the render time is 30 ms, but if I render them grouped by textures, it's 10 ms. I would never think that there's so huge difference. But I need another method to make it even faster, apart from texture atlas. I know there is Bindless Graphics extension from nVidia, but it cannot be used with AMD VGAs. Is there any Bindless Graphics like thing that can be used with AMD VGAs? The best would be independent solution, but I don't know if it exists.