GPUStack, Padler and llama.cpp

[git2212]

I was looking over GPUStack and Padler for a multi node backend using llama.cpp. I am curious if anyone had any experience with either of these two?
Like any distributed framework some latency would be expected, but how much?

Thx.