OpenAI Releases MRC Open Networking Protocol for AI Supercomputer Clusters

OpenAI released MRC (Multipath Reliable Connection), an open networking protocol developed with AMD, Broadcom, Intel, Microsoft, and NVIDIA for large-scale AI training infrastructure. MRC spreads packets across hundreds of paths simultaneously and recovers from network failures in microseconds. The protocol enables supercomputers with over 100,000 GPUs to be built using only two tiers of Ethernet switches, improving GPU networking performance and resilience.

MRC protocol capabilities

• ▪MRC (Multipath Reliable Connection) improves GPU networking performance and resilience in large-scale AI training clusters.
• ▪MRC (Multipath Reliable Connection) spreads packets across hundreds of paths simultaneously.
• ▪MRC (Multipath Reliable Connection) recovers from network failures in microseconds.
• ▪MRC (Multipath Reliable Connection) enables supercomputers with over 100,000 GPUs to be built using only two tiers of Ethernet switches.

Industry partnership

• ▪OpenAI released MRC (Multipath Reliable Connection) as an open networking protocol.
• ▪OpenAI developed MRC (Multipath Reliable Connection) in partnership with AMD, Broadcom, Intel, Microsoft, and NVIDIA.

1 source