>>62506446Why buy when you can rent? I love it when they post "1024x H100 have trained the whole stable diffusion dataset in 2.5 minutes!"
Okay, and what am I doing now with the dead weight of 1024 gpus waiting for a new dataset? There's no way somebody can even download a new dataset of a few million images in 2.5 minutes, less so program the whole thing for a new training run in that time. So they're just idling for no reason.
I get your joke though. What Pippa lora are you using? The one I have always makes crazy murderous eyes and I was thinking about retraining.