You can take a peek at the WPF architecture to get an idea of how it's set up.
To answer your specific question: WPF uses a retained mode drawing system. In particular, that means that any redrawing necessary (which may be necessary) is handled behind the scenes without your intervention.
By contrast, GDI uses immediate mode; i.e. you essentially write pixels directly and if -for any reason- those pixels need refreshing, you need to re-render.
WPF doesn't necessarily actually cache everything - that depends on the OS and memory availability amongst other things. However, if it does need to re-render, it'll use the scene graph you last provided to do so; it's transparent to the programmer. Also, even though it "uses" DirectX, that basically just means it's doing a best-effort case to use hardware support insofar available and implemented for that feature. Not all graphics cards nor all WPF features are fully accelerated. This question concerns telling the different rending modes apart and the consequence that entails.