I am getting better drawing performance when using bigger textures. A very small texture (64x64) magnified on a planer rect around 480x480 is slower then using a 512x512 texture.
The magnification filter is set to GL_NEAREST.
To be more specific on the context:
I am copying YUV video frames into the texture. A fragment shader is transforming the YUV colors into rgba. As glTexSubImage2D with GL_LUMINANCE gave me bad performance at the copy-stage, I switched to GL_RGBA, which is very fast copied compared to GL_LUMINANCE. Now I am packing 4 y-pixels into on rgba-pixel. So the texture size got four times smaller as before. And now, the drawing performace is worse as before. I was thinking, it is because of the raised shader complexity. But after optimizing the shader code (which is smaller and faster as the one I was using with GL_LUMINANCE), there was no performance gain. I am getting better drawing performance when using bigger video sizes and due to a bigger texture size. Sure, the video decompression takes more time, but performance counting just the drawing part shows much better results. But it is not the texture magnification filter which should be used more often when drawing small textures into bigger rectangles. This is set to GL_NEAREST. I looks, like re-reading pixel values more often from the same texture location makes things worse. Why?