FlashNEWS BRIEFING

OpenAI Opens Source MRC Protocol for Large-Scale AI Training Networks

1 min read

88Z AI, 7 May — OpenAI, together with AMD, Broadcom, Intel, Microsoft, and NVIDIA, has launched the Multi-path Reliable Connection (MRC) protocol, open-sourced to the industry via OCP. Built on RoCE extensions, MRC uses multi-plane network design and adaptive packet spraying to distribute data across hundreds of parallel paths. This approach prevents congestion and hotspots, reducing fault recovery time from seconds to microseconds.

At a glance

  • OpenAI and five major partners open-source MRC protocol via OCP
  • Multi-path parallel transmission reduces fault recovery to microseconds
Cover image for OpenAI Opens Source MRC Protocol for Large-Scale AI Training Networks
Cover image for OpenAI Opens Source MRC Protocol for Large-Scale AI Training Networks

The update

OpenAI has partnered with five major tech giants—AMD, Broadcom, Intel, Microsoft, and NVIDIA—to introduce the Multi-path Reliable Connection (MRC) protocol, which is being open-sourced to the entire industry through the Open Compute Project (OCP). This initiative aims to address network bottlenecks in ultra-large-scale AI training. The MRC protocol extends Remote Direct Memory Access (RoCE) by innovatively employing a multi-plane network architecture combined with adaptive packet spraying technology. By intelligently distributing data traffic across hundreds of parallel transmission paths, the protocol effectively avoids network congestion and hotspots. Its significant advantage lies in high reliability, drastically reducing fault recovery time from the traditional second-level range to microseconds, thereby greatly enhancing the training efficiency and stability of large-scale clusters.

Sources & editorial note

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.

  1. 量子位Source published

Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.

Related reading

All stories
88Z.ai