accelerated 3D, while it isn't as correct, looks impressive and convincing to a point by rendering an insane number of polygons - it's well suited for for solar, city, and geo mapping. ultimately it depends on "what you want to see". were i to choose 1 method i would choose it, i learned the language in 1997 i think, but raytracing i did mid 80's?
keep in mind older "3D accelerated" code uses "fake shadows", newer use a combination of shadowing build-in to the video card features and pre-rendered textures (sometimes raytraced textures). because many of the "documents" on the subject are by gamers you can get misled upon what these cards have in silicon and what must still be "faked" (they have hardware shadows and light sources these days, as well as other effect that aid in "hiding the crime" of faked 3D depth clues) and how (xbox, or city simulations) develope large scenes in production. But mirrors must be faked (render scene backwards, past texture on the mirror), where as in raytracing as you know the reflectivity is specified, and the rays do rest.
raytracing can, if used properly, make images that look as if they are real: but can take a week or a month to render. it's "needed" but less needed than GL?
i don't think you'll enjoy studying GL as much as geology or even lightwaves because you seem inclined to science moreso than "short code tricks that fake depth clues at high speed" and "massive code bases that assembled textures and scenery"
