GPUs with 4GB is above can be use, although having 16GB is recommended.
A good indication of VRAM requried for Llama 3.1 (one of the models we support) is available here: