I'm new in Microsoft CNTK
Google Tensorflow used grpc for communicating with each machine , but i don't know which pipeline used for distributed System in CNTK, Can you let me know?
and Could you give me some reference or site about example of multi machines , multi GPU? i already have been to this site , but i cannot find multi machine information or code.
THANK YOU! :)
This has changed over various versions but the most current information is that the binary releases under Linux and GPU are built with Nvidia's NCCL 2 library (since version 2.3) and under Windows they are built with MS-MPI. Other versions are built with OpenMPI.
The page you are linking to has a lot of information especially in the sections about the necessary changes to your Python script (they are minimal) and how to invoke it.