PVM assigns an architecture name to each kind of machine on which it runs, to distinguish between machines that run different executables.
This allows PVM to select the correct executable at run time. It also allows builds of the system or applications to proceed in parallel (in separate directories named after the architecture).
PVM also assigns a data encoding to each machine; One or more architectures map to each data encoding. Machines with different architectures, but the same data encoding must run different executables, but can pass data back and forth without converting it.