If you remember back in the day when we just had MSAA and nothing else, it was horribly slow. “Just render more pixels” turns out to give you poor efficiency. B

If you remember back in the day when we just had MSAA and nothing else, it was horribly slow. “Just render more pixels” turns out to give you poor efficiency. Better efficiency = you can spend the cycles doing something more interesting. In other words—if GPUs were powerful enough to just push more pixels to the screen, then we would still use these upsampling techniques anyway, save a bunch of computational power, and then spend that computational power improving the graphics in other ways.