Aurora will provide users with MPI implementations from both Intel and Cray, both of which will be based on the latest Argonne MPICH code base and compliant with the MPI 3.1 standard.The Aurora MPI implementations will be optimized for the Aurora network topology, for the on-node interconnects, and for the Intel GPUs.

In addition, the MPI librarieswill be thread safe and allow applications to use MPI from individual threads. Asynchronous progress in all types of nonblocking communicationincluding nonblocking send-receive, nonblocking collectives and One-sidedoperations will besupportedand optimizations forhardware-offloading to the NIC and switchwill also be provided.