How does "deferred shading" address the challenge of rendering scenes that contain many dynamic light sources?

A It removes the need for a depth buffer entirely, saving memory B It always produces lower visual quality than forward rendering, in exchange for speed C It compresses vertex data prior to the rasterization stage D It first writes geometry attributes (depth, normals, albedo) into G-buffers, then computes lighting in a separate pass — avoiding redundant shading of occluded fragments and scaling well with many lights