Comment by mechagodzilla

I have a used workstation I got for $2k (with 768GB of RAM) - using the Q4 model, I can get about 1.5 tokens/sec and use very large contexts. It's pretty awesome to be able to run it at home.