Computer Graphics can be divided in multiple subdomains of which I will only talk about physically-based rendering (the one I am the most familiar with and probably the one you are referring to based on the tags of the question).
Physically-based rendering (also referred to as global illumination) is generally far from real time at the moment. The speed we achieve is in fact no way close to what we actually try to simulate: the propagation of light through scenes (see interview with prof. Ravi Ramamoorthi). Scenes are rendered for over multiple hours or even days, far below the actual speed of light. Current research in this domain does however not aim at these speeds, a non-flickering framerate perceived by a human observer will do. Since the equations we try to solve are defined a long time ago, effort is mostly put in optimizing current algorithms which can be at a very high (introducing new formalisms) or low level (changing the right 1s to 0s and vice versa).
For the ones who actually want to dive in the literature, see this wonderful up-to-date site of the papers accepted at the most important conferences in the field of computer graphics.