Multiple GPUs / Sampler Method

[Dan-Burns]

I'm curious about how parallel simulations are run.

For replica exchange, will the systems be divided among available GPUs if there are fewer GPUs than systems? Can you run across multiple nodes?

Also what's the recommendation for dealing with time limits. If I can't converge in a time-limited run and there is no checkpoint option, is there a way to set the coordinates of each state to the ending coordinates of the first run, run again, and append results?