But slowly, I filled my home server with whatever CUDA capable cards I have and it’s fine for SD, but I found llama way too slow. I rented a dual A2000 instance for a couple weeks and it was bearable, but still not great.