It uses both CPU and GPU. There are a lot of factors. Starting with main board’s Bus speed. This is the fastest that data can be transferred between RAM, CPU, and GPU. Then you have RAM’s data transfer rate. Then there’s the CPU’s FSB(Front Side Bus). Ideally you want a FSB that matches the Bus speed. Next you have the CPU’s cache. Level 3 is the CPU’s memory for holding data until it gets assigned a core. Level 2 is holding for spot per core. Level 1 is the data that is being processed. Some processors list all 3 as total cache and some have each listed separately. The Level 1 and Level 2 get multiplied by the number of cores when comparing CPUs. On a GPU the VRAM is the GPU’s cache. That is why a GPU can process much faster than a CPU. Both CPU and GPU have to transfer data to and from the RAM in most cases. I’m not sure on the newer GPU/CPU combo chips(APU I think is what they are called or that may only be what AMD calls them). In my limited experience, the main board(Bus) and RAM(Data Transfer) speeds are usually where the bottle neck happen.
