英伟达推出了 Minerva，这是一款 AI 运维软件，它将 GPU、网络、存储和工作负载的数据结合起来，以便在大型 AI 数据中心中及早发现故障部件。它将警报归类为单一事件，并配备了 AI 助手。IREN 正在使用它，Lambda、IBM 和特斯拉计划使用。
Nvidia has launched Minerva, AI operations software that combines data from GPUs, networks, storage and workloads to spot failing parts in large AI data centres early. It groups alerts into single incidents and has an AI assistant. IREN uses it, and Lambda, IBM and Tesla plan to.
英伟达周四表示，该公司推出了 Minerva，这款软件监控其称为“AI 工厂”的巨大 GPU 集群，并在问题拖慢工作进度前标记出来。该软件在英伟达 GTC 柏林大会上发布，此前几天英伟达刚与德国签署 AI 合作伙伴关系。
Nvidia has launched Minerva, software that watches over the huge GPU clusters it calls AI factories and flags problems before they slow work down, the company said on Thursday. It was announced at Nvidia’s GTC Berlin conference, days after Nvidia signed an AI partnership with Germany.
英伟达称，在如此规模的集群中，一个故障部件可能会拖垮整个训练任务。可能是某个 GPU 比其他 GPU 运行温度更高，或者存储驱动器在负载下变慢。当网络、计算和机器学习团队各自监控系统的不同部分时，这些迹象很容易被忽略。
In clusters that size, one faulty part can hold up an entire training job, Nvidia says. It might be a GPU running hotter than the rest, or a storage drive slowing under load. Those signs are easy to miss when network, computing and machine learning teams each watch a separate part of the system.
Minerva 将来自 GPU、网络设备、存储、作业调度器和 AI 工作负载本身的数据汇聚到一张数据中心地图中。英伟达表示，这使运维人员能追溯缓慢任务背后的致因部件，即使跨越数十万个 GPU 也能做到。
Minerva pulls data from GPUs, network gear, storage, job schedulers and the AI workloads themselves into one map of the data centre. Nvidia says that lets operators trace a slow job back to the part causing it, even across hundreds of thousands of GPUs.
该软件利用机器学习来了解正常状态的样子并预测故障。它不向运维人员发送每一个警报，而是将相关问题归类为单一事件，并给出可能的原因。
The software uses machine learning to learn what normal looks like and to predict failures. Instead of sending operators every alert, it groups related problems into a single incident with a likely cause.
内置的 AI 助手让员工能用自然语言询问集群健康状况，并提出修复建议。团队可以手动应用这些修复，或将 Minerva 接入自己的自动化工具。
A built-in AI assistant lets staff ask about cluster health in plain language and suggests fixes. Teams can apply those fixes by hand or hook Minerva up to their own automation tools.
“NVIDIA Minerva 让我们的运维团队能在一个视图中看到网络、GPU 以及运行在其上的作业，”IREN 首席技术官 Denis Skrinnikoff 表示。根据 5 月达成的协议，英伟达最多可向这家前比特币矿企投资 21 亿美元。
“NVIDIA Minerva gives our operations team one view across the network, the GPUs and the jobs running on them,” said Denis Skrinnikoff, chief technology officer of IREN. Nvidia can invest up to $2.1B in the former Bitcoin miner under a deal struck in May.
英伟达表示，Bitdeer AI、IBM、Jio、Lambda 和特斯拉也计划使用 Minerva。戴尔、联想和超微电脑将向其客户提供该软件。
Bitdeer AI, IBM, Jio, Lambda and Tesla also plan to use Minerva, Nvidia said. Dell, Lenovo and Supermicro will offer it to their customers.
此次发布进一步推动了英伟达向构建 AI 数据中心的公司销售除芯片以外产品的势头。周三，该公司表示欧洲合作伙伴正在采用其 DSX 蓝图以削减水电消耗，并发布了一项用于安全运行 AI 智能体的设计。英伟达未公布 Minerva 的定价。
The launch adds to Nvidia’s push to sell more than chips to the companies building AI data centres. On Wednesday, it said European partners are adopting its DSX blueprint to cut water and power use, and released a design for running AI agents safely. Nvidia did not give pricing for Minerva.