Basically need VMs with 8gb-24gb gpu attached so I can run my own models, spin up and spin down via API, bill per second?