4 ms·

I have a multi-GPU rig designed exactly for this purpose :) Check out r/localllama. There are literally dozens of us!