There are different ways to light up the scene; some techniques are static and pre-calculated and some are dynamic and real-time.

No single technique for reflections suffices. Unreal Engine uses a combination of 3:

Unreal Engine always compresses the textures that you import. Some compression methods do a better job at maintaining quality while reducing the size. Other methods serve special purposes: DXT5/BC5, for example, is used for compressing normal maps; they only store red and green, and blue can then be derived from them (that is, z can be derived from x and y). BC stands for block compression.

In the Materials editor, you can see the HLSL code of a material.

The pixel shader executes mesh by mesh, because a pixel shader execution is the result of a draw call, and a draw call (or more) is made per mesh.

Early Z-pass is a kind of object depth testing for avoiding the execution of the pixel shader on objects whose pixels will be overwritten by the pixels of objects in front of them.

Much in rendering is a mix of techniques used simultaneously. A single one doesn’t cut it. That is true for occlusion.

The basic performance metrics are frame rate, and its inverse, milliseconds per frame.

You can set up Unreal to do either deferred or forward rendering.

Ray Tracing in One Weekend by Peter Shirley must be the best end-to-end introduction to ray tracing out there: it’s accesible enough for beginners, but sufficiently rigorous to lead to a correct implementation.