The base primitive a GPU draws is triangles; they are the fastest primitive for math-y reasons, but the basic premise is that you can subdivide any triangle into two right-angle triangles, which have favorable properties for scanline-based rasterization.
Now, software modelling packages can represent geometry in any way they want, as long as that software package can somehow eventually transform that representation into triangles for the GPU to rasterize to pixels (barring custom software rasterizers).
My comment referred to quads, but at the end of the day, a quad is 2 triangles. Same goes for arbitrary polygons. The GPU speaks triangles.
Now, software modelling packages can represent geometry in any way they want, as long as that software package can somehow eventually transform that representation into triangles for the GPU to rasterize to pixels (barring custom software rasterizers).
My comment referred to quads, but at the end of the day, a quad is 2 triangles. Same goes for arbitrary polygons. The GPU speaks triangles.