--- title: MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet description: Meta introduces MetaRoCE, a clean-sheet RDMA transport protocol designed for AI workloads on commodity Ethernet at million-GPU scale. Unlike standard RoCE,... canonical: https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet | daily.dev og:description: Meta introduces MetaRoCE, a clean-sheet RDMA transport protocol designed for AI workloads on commodity Ethernet at million-GPU scale. Unlike standard RoCE,... og:url: https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4 og:image: https://api.daily.dev/og/posts/4RzMHAXu4.png og:image:alt: MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet **[Facebook Engineering ](https://daily.dev/sources/facebook_code)** · 8 min read · 0 upvotes · 0 comments ## Summary Meta introduces MetaRoCE, a clean-sheet RDMA transport protocol designed for AI workloads on commodity Ethernet at million-GPU scale. Unlike standard RoCE, which relies on in-order delivery and PFC-based losslessness, MetaRoCE moves intelligence to the NIC, embracing native out-of-order delivery, multipathing with per-path telemetry, loss tolerance without pause frames, and dual-sided congestion control. Tested on a 64-node AMD GPU cluster with Pensando NICs, it showed higher throughput and lower flow completion times than RoCEv2, maintaining ~86% throughput at 1% packet loss and degrading gracefully even at 10% loss. Meta is contributing the specification, a compliance suite, and a software reference implementation (libsoftmetaroce) to the Open Compute Project, with a full release planned at the 2026 OCP Global Summit in October. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is MetaRoCE and how does it differ from standard RoCEv2? MetaRoCE is a clean-sheet RDMA transport protocol built by Meta for AI workloads on commodity Ethernet. Unlike standard RoCEv2, which requires in-order packet delivery and PFC-based losslessness, MetaRoCE tolerates loss and out-of-order arrival by design, sprays packets across many per-connection paths with individual telemetry, and moves congestion and ordering intelligence to the NIC instead of the switch fabric. _Track emerging RDMA transport designs like MetaRoCE as AI network architectures evolve, via daily.dev._ ### How does MetaRoCE perform under packet loss compared to RoCEv2? MetaRoCE maintains roughly 86% throughput at 1% packet loss and continues delivering useful bandwidth even at 10% loss, converging gracefully instead of collapsing. Tests ran on a 64-node AMD GPU cluster using RCCL collectives with AMD Pensando programmable NICs, comparing all-reduce and all-to-all operations against RoCEv2, which degrades sharply under similar loss conditions. _Compare network transport resilience benchmarks like these on daily.dev when evaluating AI infrastructure choices._ ### When will Meta release the MetaRoCE specification and reference implementation publicly? Meta plans to release the MetaRoCE specification, a DPDK-optimized software reference implementation, and a production compliance framework in October at the 2026 OCP Global Summit. The specification is being contributed to the Open Compute Project for multi-vendor adoption, alongside the libsoftmetaroce software reference stack that runs on commodity Linux over standard UDP sockets. _Follow open networking standard releases such as this one on daily.dev to plan infrastructure adoption timelines._ ## Similar posts on daily.dev - [What is RDMA over Converged Ethernet \(RoCE\)?](https://daily.dev/posts/what-is-rdma-over-converged-ethernet-roce--rjpagkmdv) · Ubuntu · 0 upvotes · 0 comments - [NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC](https://daily.dev/posts/nvidia-spectrum-x-the-open-ai-native-ethernet-fabric-sets-the-standard-for-gigascale-ai-now-wi-8nauo5ewr) · NVIDIA · 0 upvotes · 0 comments - [The Counterintuitive Networking Decisions Behind OpenAI’s 131,000-GPU Training Fabric](https://daily.dev/posts/the-counterintuitive-networking-decisions-behind-openai-s-131-000-gpu-training-fabric-zn8imcu8v) · Towards Data Science · 0 upvotes · 0 comments --- Tags: [#networking](https://daily.dev/tags/networking), [#ai-infrastructure](https://daily.dev/tags/ai-infrastructure) [View this post on daily.dev](https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet","url":"https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4"},"datePublished":"2026-08-24T18:05:38.872Z","dateModified":"2026-08-24T18:06:12.994Z","description":"Meta introduces MetaRoCE, a clean-sheet RDMA transport protocol designed for AI workloads on commodity Ethernet at million-GPU scale. Unlike standard RoCE,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/817f4a93eae01e94c1abf56ec4620fbb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/817f4a93eae01e94c1abf56ec4620fbb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Facebook Engineering\n","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Facebook Engineering\n","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/facebook","url":"https://daily.dev/sources/facebook_code"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"networking,ai-infrastructure","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Facebook Engineering\n","item":"https://daily.dev/sources/facebook_code"},{"@type":"ListItem","position":3,"name":"MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4#faq","mainEntity":[{"@type":"Question","name":"What is MetaRoCE and how does it differ from standard RoCEv2?","acceptedAnswer":{"@type":"Answer","text":"MetaRoCE is a clean-sheet RDMA transport protocol built by Meta for AI workloads on commodity Ethernet. Unlike standard RoCEv2, which requires in-order packet delivery and PFC-based losslessness, MetaRoCE tolerates loss and out-of-order arrival by design, sprays packets across many per-connection paths with individual telemetry, and moves congestion and ordering intelligence to the NIC instead of the switch fabric. Track emerging RDMA transport designs like MetaRoCE as AI network architectures evolve, via daily.dev."}},{"@type":"Question","name":"How does MetaRoCE perform under packet loss compared to RoCEv2?","acceptedAnswer":{"@type":"Answer","text":"MetaRoCE maintains roughly 86% throughput at 1% packet loss and continues delivering useful bandwidth even at 10% loss, converging gracefully instead of collapsing. Tests ran on a 64-node AMD GPU cluster using RCCL collectives with AMD Pensando programmable NICs, comparing all-reduce and all-to-all operations against RoCEv2, which degrades sharply under similar loss conditions. Compare network transport resilience benchmarks like these on daily.dev when evaluating AI infrastructure choices."}},{"@type":"Question","name":"When will Meta release the MetaRoCE specification and reference implementation publicly?","acceptedAnswer":{"@type":"Answer","text":"Meta plans to release the MetaRoCE specification, a DPDK-optimized software reference implementation, and a production compliance framework in October at the 2026 OCP Global Summit. The specification is being contributed to the Open Compute Project for multi-vendor adoption, alongside the libsoftmetaroce software reference stack that runs on commodity Linux over standard UDP sockets. Follow open networking standard releases such as this one on daily.dev to plan infrastructure adoption timelines."}}]} ```