OpenAI Got The Whole AI Squad To Accelerate Large-Scale AI Training – AMD, NVIDIA, Intel, Microsoft & Broadcom All-In On MRC

OpenAI and partners unveiled MRC, a new multipath networking protocol for large-scale AI training, released through the Open Compute Project.

MRC splits 800 Gb/s links into multiple paths to improve resilience, cluster efficiency, and faster recovery from network failures.

The protocol was developed over two years with major chip and cloud companies, including AMD, NVIDIA, Intel, Microsoft, and Broadcom.

MRC is already deployed in OpenAI supercomputers using NVIDIA and Broadcom hardware, with aims to simplify scaling across very large GPU clusters.