OpenAI Introduces MRC (Multipath Reliable Connection): A New Open Networking Protocol for Large-Scale AI Supercomputer Training Clusters

OpenAI released MRC, an open networking protocol for large AI training clusters, through the Open Compute Project.

MRC extends RoCE with packet spraying and SRv6-based source routing to spread traffic across many paths.

The design aims to recover from network failures in microseconds and keep GPUs from sitting idle during congestion.

OpenAI built the protocol with major chip and cloud partners, including AMD, Broadcom, Intel, Microsoft, and NVIDIA.