Skip to content
AI News HubLIVE
More
Summary only

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

OpenAI, in partnership with AMD, Broadcom, Intel, Microsoft, and NVIDIA, has released MRC, an open networking protocol for large-scale AI training clusters. MRC uses adaptive packet spraying to eliminate congestion, achieves microsecond-level failure recovery via static source routing, and enables two-tier network topologies supporting over 100,000 GPUs. Already deployed in production across OpenAI's largest supercomputers, MRC has been used to train frontier models for ChatGPT and Codex.

SourceMarkTechPostAuthor: Michal Sutter
OpenAI Introduces MRC (Multipath Reliable Connection): A New Open Networking Protocol for Large-Scale AI Supercomputer Training Clusters
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • OpenAI, AMD, Broadcom, Intel, Microsoft, and NVIDIA co-developed MRC, released via Open Compute Project.
  • MRC uses intelligent packet spraying across hundreds of paths to reduce congestion and tail latency.
  • Microsecond failure recovery keeps training jobs alive during link or switch failures.
  • Two-tier multi-plane network supports 131,000 GPUs with fewer switches and optics, reducing cost and latency.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models