[D] Cheapest way to scale up and down from a large GPU to just CPU with minimal overhead?

I've got a bunch of hobbyist projects to work on while I'm away from home for a few weeks and away from my fairly powerful GPU desktop.

I'm looking to find the simplest (and cheapest) way to get a computer that I can SSH into and:

1. Do some coding/debugging (spend 80% of my time)
2. Run some fairly niche ML software in essentially batch mode that needs a low end GPU (10%)
3. Fire up a powerful 48 GB …

computer cpu desktop gpu home machinelearning projects scale work