Graphics cards were created to render graphics in a certain way and at "realtime" speeds - 30+ frames per second. Because they have to render at that speed they use many shortcuts, workarounds, and lower quality rendering methods that are faster. Newer graphics cards can achieve better and better quality and the rendering approach used by games changes somewhat over time, but it is still far from generalized raytracing, global illumination and other more realistic rendering techniques that are used in renderers like Mental Ray (used by 3DS Max, Maya and Softimage).
So the basic answer is that graphics cards do accelerate a *type* of rendering, but they cannot (yet) effectively process the kind of rendering approach that creates high quality results like Terragen 2, Mental Ray, etc. As graphics cards continue to develop this may become more feasible, but we are also seeing the introduction of more and more CPU cores, so by the time graphics cards are capable enough to accelerate general raytracing, we may see CPU's with upwards of 64 cores, which might keep CPU's ahead of graphics processors for that kind of rendering still.
In any case it's a ways off and for now TG2's CPU-based approach is the most effective approach, and the one that the vast majority of professional renderers use as well. Even large companies like Autodesk, who have tremendous resources to put on the task of accelerating normal rendering with graphics cards, still rely on CPU-based rendering, which shows how generally infeasible it is with present hardware.
- Oshyan