tf slim - distributed tensorflow clarification -

is understanding correct model_deploy lets user train model using multiple devices on single machine? basic premise seems clone devices variable sharing , variables distributed param servers in round-robin fashion.

on other hand distributed tensorflow framework enables user train model through cluster. cluster lets user train model using multiple devices across multiple servers.

i think slim documentation slim , point has been raised couple of times already: configuration/flags tf-slim across multiple gpu/machines

thank you.