MetaRoCE：一种专为AI-Scale Ethernet构建的全新RDMA传输协议

Training and serving frontier AI models depends on fast, reliable networks that move data between GPUs without wasting compute cycles.
训练和服务前沿AI模型依赖于快速、可靠的网络，这些网络在GPU之间传输数据而不会浪费计算周期。
To meet this challenge at scale, Meta designed MetaRoCE – a clean-sheet RDMA transport protocol purpose-built for AI workloads on commodity Ethernet.
为了在大规模下应对这一挑战，Meta 设计了 MetaRoCE——一种专为通用以太网上的 AI 工作负载从头设计的 RDMA 传输协议。
We’re releasing the MetaRoCE specification, a reference software implementation and a compliance test suite through the Open Compute Project (OCP) to enable the broader industry to adopt, implement, and build on it.
我们正在通过 Open Compute Project (OCP) 发布 MetaRoCE 规范、参考软件实现和合规性测试套件，以使更广泛的行业能够采用、实现并在此基础上进行构建。

At Meta, we’ve been a strong driver behind the industry’s growing consensus that Ethernet should be the fabric of choice for AI infrastructure. We’ve already shown that RoCE can power distributed AI training at scale. Now, we’re building on that work with MetaRoCE, protocol designed from the ground up for Ethernet at million-GPU scale.

在 Meta，我们一直是推动行业日益形成的共识——即 Ethernet 应成为 AI 基础设施的首选 fabric 的强大驱动力。我们已经证明 RoCE 能够大规模支持分布式 AI 训练。现在，我们正在通过 MetaRoCE 在此基础上继续发展，这是一种从零开始为百万 GPU 规模的 Ethernet 设计的协议。

We’ve scaled up clusters of hundreds of thousands of GPUs, spread over multiple data centers and regions. Whether these clusters are training the next frontier model or serving inference at global scale, the network is in the critical path.

我们已经将包含数十万个 GPU 的集群进行了扩展，分布在多个数据中心和区域。无论这些集群是在训练下一个前沿模型，还是在全球范围内提供推理服务，网络都处于关键路径中。

Collective operations like all-reduce and all-to-all synchronize thousands of accelerators during training, and the slowest transfer sets the pace for the entire job. In inference, low-latency communication between distributed model shards directly impacts response times for hundreds of millions of users. Even small amounts of network friction directly strand significant compute capacity.

像 all-reduce 和 all-to-all 这样的集合操作在训练期间同步数千个加速器，最慢的传输决定了整个任务的速度。在推理中，分布式模型分片之间的低延迟通信直接影响数亿用户的响应时间。即使是少量的网络摩擦也会直接导致大量计算能力闲置。

Stan...