<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[OpenAI 联合英伟达等五家公司发布 MRC 网络协议，解决 GPU 集群拥塞与故障问题并开源]]></title><description><![CDATA[<p dir="auto">OpenAI、微软、英伟达、AMD、英特尔及博通的研究团队于 5 月 6 日联合发布论文，正式推出历时两年研发的数据中心网络传输协议 MRC（多路径可靠连接，Multipath Reliable Connection），并通过开放计算项目（OCP）以开放许可证发布规范。MRC 针对大规模 GPU 集群中两大核心痛点而设计：一是拥塞——采用&quot;数据包喷射&quot;（packet spraying）机制，将数据流同时分散至数百条网络路径，避免单链路过载，同时压平集群网络层级，降低整体能耗；二是故障——可在微秒级内检测路径中断并自动切换，使训练任务在网络局部故障时无需中断。此外，MRC 还与 SRv6（IPv6 段路由）协议配合，由发送端直接规划数据路径，大幅减少交换机的路由计算负担。OpenAI 网络负责人 Mark Handley 表示，MRC 让集群网络从&quot;故障放大器&quot;变为弹性基础设施。</p>
<p dir="auto">MRC 目前已部署于 OpenAI 与微软最大的训练集群，包括甲骨文德克萨斯州 Abilene 站点及微软的 Fairwater 超算，并已参与训练多个 OpenAI 模型。英伟达方面同步宣布，Spectrum-X 以太网平台原生支持 MRC 协议，同时兼容此前的自适应 RDMA 方案，为客户提供传输层选择弹性。OpenAI 工作负载负责人 Greg Steinbrecher 强调，推动 MRC 开源的出发点是消除行业碎片化——当前各家大型公司各自维护私有协议，整体效率低下，统一标准后算力资源才能得到更充分利用。</p>
<p dir="auto"><a href="https://blogs.nvidia.com/blog/spectrum-x-ethernet-mrc/" target="_blank" rel="noopener noreferrer nofollow ugc">NVIDIA Blog</a> | <a href="https://www.thedeepview.com/articles/exclusive-openai-unveils-protocol-to-stretch-compute" target="_blank" rel="noopener noreferrer nofollow ugc">The Deep View</a></p>
<p dir="auto"><div class="card col-md-9 col-lg-6 position-relative link-preview p-0">



<a href="https://blogs.nvidia.com/blog/spectrum-x-ethernet-mrc/" title="NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC">
<img src="https://blogs.nvidia.com/wp-content/uploads/2026/05/ethernet-press-spectrum-x-corp-blog-1920x1080-1.jpg" class="card-img-top not-responsive" style="max-height: 15rem;" alt="Link Preview Image" onerror="this.parentElement.remove()" />
</a>



<div class="card-body">
<h5 class="card-title">
<a class="text-decoration-none" href="https://blogs.nvidia.com/blog/spectrum-x-ethernet-mrc/">
NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC
</a>
</h5>
<p class="card-text line-clamp-3">Multipath Reliable Connection — a new transport protocol proven first and optimized on NVIDIA Spectrum-X Ethernet hardware — is now open to the industry.</p>
</div>
<a href="https://blogs.nvidia.com/blog/spectrum-x-ethernet-mrc/" class="card-footer text-body-secondary small d-flex gap-2 align-items-center lh-2">



<img src="https://blogs.nvidia.com/favicon.ico" alt="favicon" class="not-responsive overflow-hiddden" style="max-width: 21px; max-height: 21px;" onerror="this.remove()"/>



<p class="d-inline-block text-truncate mb-0">NVIDIA Blog <span class="text-secondary">(blogs.nvidia.com)</span></p>
</a>
</div></p>
<p dir="auto"><div class="card col-md-9 col-lg-6 position-relative link-preview p-0">



<a href="https://www.thedeepview.com/articles/exclusive-openai-unveils-protocol-to-stretch-compute" title="Exclusive: OpenAI unveils protocol to stretch compute">
<img src="https://cdn.sanity.io/images/0no9te5e/production/efd7c865f9a8a0a7ac530b60e35224a96faf0d10-1456x816.png?rect=0,28,1456,761&w=1200&h=627&fit=crop&auto=format" class="card-img-top not-responsive" style="max-height: 15rem;" alt="Link Preview Image" onerror="this.parentElement.remove()" />
</a>



<div class="card-body">
<h5 class="card-title">
<a class="text-decoration-none" href="https://www.thedeepview.com/articles/exclusive-openai-unveils-protocol-to-stretch-compute">
Exclusive: OpenAI unveils protocol to stretch compute
</a>
</h5>
<p class="card-text line-clamp-3">OpenAI is getting creative to deal with the industry's imminent compute crunch.

On Wednesday, the ChatGPT maker and a coalition of researchers from A...</p>
</div>
<a href="https://www.thedeepview.com/articles/exclusive-openai-unveils-protocol-to-stretch-compute" class="card-footer text-body-secondary small d-flex gap-2 align-items-center lh-2">



<img src="https://www.thedeepview.com/favicon.ico" alt="favicon" class="not-responsive overflow-hiddden" style="max-width: 21px; max-height: 21px;" onerror="this.remove()"/>





<p class="d-inline-block text-truncate mb-0">The Deep View <span class="text-secondary">(www.thedeepview.com)</span></p>
</a>
</div></p>
]]></description><link>https://welinux.com//topic/211/openai-联合英伟达等五家公司发布-mrc-网络协议-解决-gpu-集群拥塞与故障问题并开源</link><generator>RSS for Node</generator><lastBuildDate>Mon, 18 May 2026 21:15:27 GMT</lastBuildDate><atom:link href="https://welinux.com//topic/211.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 06 May 2026 14:46:01 GMT</pubDate><ttl>60</ttl></channel></rss>