A thousand chips train the same model, and the whole time they tell each other what they computed. Lose one packet from that conversation and the standard protocol stops hurrying.
MetaRoCE is Meta's answer. Not a patch on the old RoCE, but a protocol written from a blank page, specifically for the collective operations between thousands of accelerators, the conversations where everyone waits for everyone.
Fact
The facts: MetaRoCE is a new transport protocol for direct memory-to-memory transfer, written from scratch for AI workloads over plain Ethernet. By Meta's own numbers, it keeps about 86 percent of throughput at 1 percent packet loss, with higher throughput and shorter flow completion time than RoCEv2. It scales linearly on multi-plane networks, tested on 4 and on 8 planes with 4,000 simultaneous connections. It is opening through the Open Compute Project with a full spec, a software reference implementation named libsoftmetaroce, and a compliance test suite. Among the deployment partners named is AMD Pensando. Source: Meta's engineering blog.
Why put it out in the open
A network protocol written inside one company usually stays there. Meta does the opposite and puts the spec, the code and the test into the Open Compute Project, where any chipmaker can check its own silicon against them.
At one percent packet loss, most protocols choke. This one, by Meta's own numbers, keeps about 86 percent of speed.
The math is simple. An open protocol only wins if enough chips speak it. AMD Pensando is the first partner named. If three big chipmakers speak it within a year, Meta has won itself a standard. If not, it just gave away good engineering for free.
Official primary sources
Meta Engineering: MetaRoCEThe visual is generated code-art, no third-party images.