INFRA Signal 166
Meta releases MetaRoCE RDMA transport specification and reference implementation through OCP
↗
Meta released MetaRoCE, a new RDMA transport protocol designed for AI workloads on commodity Ethernet, along with its specification, a reference software implementation, and a compliance test suite through the Open Compute Project.
Standard RoCE depends on PFC and in-order delivery, which degrade at large scale and create operational fragility. MetaRoCE eliminates PFC, embraces out-of-order delivery, and moves congestion intelligence to endpoints, which could simplify network operations for AI training clusters.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
MetaRoCE sprays packets across multiple paths and treats out-of-order arrival as normal, writing data directly to final memory locations without reorder buffers.
The protocol requires no PFC or pause frames, using per-path selective acknowledgment to retransmit exactly the missing packet on the path that lost it.
Meta released the specification, reference software implementation, and compliance test suite through the Open Compute Project for industry adoption.
THE CLUSTER