llama.cpp runs on your processor and uses a lot of RAM, so splurging on a GPU isn't going to help the performance