Powering frontier AI models like ChatGPT requires massive GPU compute working seamlessly together, with the network determining how much capacity can be utilized. To address this, AMD, in collaboration with OpenAI, Microsoft, and industry partners, has contributed Multipath Reliable Connection (MRC) to the Open Compute Project. MRC is designed for large‑scale AI training, distributing traffic across multiple paths to reduce congestion and reroute quickly during failures.
Multipath
This transforms the network into a shock absorber, ensuring consistent throughput under real‑world conditions. AMD co‑authored the MRC specification, contributed congestion control technology, and validated deployments with a leading cloud provider. Its programmable Pensando NICs enabled early implementation, now scaling to 800G. MRC redefines AI infrastructure performance by prioritizing resilience, synchronization, and efficiency over peak bandwidth. With contributions from AMD, OpenAI, Broadcom, Intel, and Microsoft, MRC advances AI networking from concept to production‑ready infrastructure, supporting cloud, enterprise, and sovereign AI environments.
Table of Contents
Read More News on Latest Malaysia
Follow us on:
Read More News on Business News Malaysia
Read More News on SG Business News
Read More News on World Future TV
Read More News #latestmalaysia



Leave a Reply
You must be logged in to post a comment.