This. No matter how complex, everything in games (and computer graphics in general) is made of triangles. You can add more triangles to make things look even more rounded, but no matter what you'll always find jagged edges if you zoom in close enough.
If you were somehow able to examine the models used in Pixar movies in real time, you'd probably still find some jagged edges if you zoomed in close enough. You'd just have to zoom in a lot more, and since CG movies are all pre-renderd, filmmakers can do whatever they want to hide any flaws that might appear.
Theoretically we could reach a point where we're using so many polygons that you'd have to zoom in to near-microscopic levels to find a jagged edge on a "round" object.
or just make sure polygons are smaller than pixels.
because we use Polygons and not Nurbs. I always wondered why Nurbs didnt take off in the gaming space, probably because of the cost, but still I expected we'd eventually see a shift.
Nurbs are cumbersome to work with, and the surface would still have to be broken down to polygons at render-time(if I'm not mistaken). Nurbs are generally used when precision is important since a nurbs surface is a mathematical equation instead of connected pre-defined points the resolution is infinite.
When you design a machine using cad software you would use something like nurbs, can't have stuff like that in an inaccurate format as polygons.
I believe you can do 3D modelling with splines, but it's pretty experimental if I remember correctly. It'd give perfect curves, however.
3d-modelling with splines has been around for ages. I remember 10 years ago or so when subd-modelling wasn't very popular, lot's of people used splines to define patch surfaces, it was a total headache compared to the tools we have today.
Also, I'm not 100% sure about this but since a spline still is only a parametrized equation at some point the computer still has to translate it into polygons to actually render it. There is no magic way of getting free perfect curves. But like I said above, if you have the computing power to spare, just make sure each polygon in a curve is smaller than a pixel on screen and the curve will always appear perfect assuming the pixels aren't big enough to see individually.
I'd say something like
open subdiv likely could find it's way into videogamin in the next 5-10 years. Its basically the tech used in cg-animation today. You start out with a lowpoly(for a high-end model that would be maybe 30 k polies for a face, but could be as low as a few thousands and still give good results) and then subdivide it with a smoothing algorithm, finally you apply a displacement map to get extra high frequency detail if you want that look. This will definitively happen when gpus are strong enough to not care about polycounts anymore.(we are approaching that point, shaders, lighting, effects and textures are far more expensive than drawing triangles). It's similar to tessellation but while tessellation just adds extra points without affecting the shape of the object subdiv changes the shape to have a curved surface.