I’ve noticed that if I take a bilinear sample of a texture, that the result is noticeably lower quality than if I read the 4 involved pixels and then do a bilinear interpolation myself in the fragment shader.
This has occured for me in both WebGL (chrome, firefox and internet explorer using windows 8) and OpenGL. I have a pair of 980m’s but also have tried other machines including a desktop.
Is there any way to get a higher quality hardware based bilinear sampling by chance, or is this a limitation of current hardware?
WebGL example at the link below. View source of the webpage to see the code: