Bairen Yi

> This is really sad. TF without MPI on clusters will use the very slow Ethernet connections You could build TF 1.x with built-in MPI support. For TF 2.x, you...

Also ping @jbedorf who is the author of the MPI plugin.

Thanks for your contribution! The performance figure looks promising. It is interesting to see some additional improvement of the original BFC allocator design, as it already does proactive registration of...