If there's not any GPU, how could I present framebuffer to the display? Or software rendering only means running graphic pipeline on CPU instead of GPU, and finally we still need a GPU to present the result?
Rendering something in software just refers to creating the framebuffer on the CPU or other programmable hardware that does not employ graphics specific circuitry. The concept is not concerned with presentation of the results, you're still doing software rendering when you write the result to disk, stream it over a network etc.