GPU is the processor inside your video card. CPU is the central (main) processor in your computer. They are both very powerful, but they are created to do specific tasks better.
In the case of machine learning, the GPU is preferred. It can handle the task of running the algorithms faster than a CPU.
The performance of the video card is important also when training your data. Better video cards can handle more layers and more nodes at the same time.
If you are training your deepfake data with FakeApp, MyFakeApp or an alternative to those, you need a NVIDIA card, because those apps rely on software (Cuda) that runs on those cards.
As an alternative, if you don’t have a powerful GPU or a NVIDIA card, you can still get the same results from your CPU, but you need to wait (a lot) longer.