AI 模型和平台
NVIDIA 发布 MinervaAIOps AI工厂可观测性平台
在 Google 上的首选来源中添加 Unite.AI
2026年10月22日的NVIDIA 发布 Minerva将其描述为AIOps平台，让运营团队提前对AI工厂的基础设施问题发出警告。 Minerva将基础设施和AI工作量的遥测合并起来，然后应用专门设计的AI模型将数据转换成可操作的洞察力，据该公司称。
AI工厂在世界上最复杂的业务中排名NVIDIA 称，拥有庞大的GPU训练模型，并服务AI跨高速网络的应用。 在这个尺度下的单个退化组件——GPU运行比它的同伴更热，或者NVMe驱动器在负载下失去性能——可以拖慢整个工作，这种警告信号很容易错过。 Minerva针对的是NVIDIA所称的AI工厂产量：基础设施通过GPU时数、良好产量、集聚异常时间和象征性生产来测量的有用工作。
工厂动态知识图
传统的监测工具往往将数据中心分割成单独的视图，网络，计算，站点可靠性和机器学习团队各自看到操作的不同部分。 NVIDIA 称 Minerva通过它的GPU，联网，存储，调度和工作量来连接这些信号。
从这些信号中，Minerva构建了一个动态知识图，从逻辑上将GPU,SuperNIC，开关和电缆与运行在它们上的工作一起绘制。 NVIDIA 称这个映射让操作员追踪一个缓慢的训练工作回到影响到它的组件，甚至在一个运行数十万GPU的工厂里。
NVIDIA认为，该平台分析时速数十亿个数据点，同时进行有监督的和无监督的机器学习，学习正常行为是什么样子，标出它以前没有遇到过的异常，在出现故障时预测故障，并利用工厂的地形将相关事件联系起来。 该平台可以将一连串相关事件的暴风雨合并为一个具有潜在根源的焦点事件，而不是将每个个体的症状传递给操作员。
AI 助理和补救工作流程
Minerva船配备了本地的AI助手，还可以与外部的代理AI工具连接，为运营商提供了一个基于直播遥测的自然语言界面。 操作员可以通过谈话调查集群健康并接受推荐的补救。 团队可以让一个人在循环中进行手动修正，或通过基础设施应用编程接口和外部系统将平台与自动化连接。
Bitdeer AI，宣布于10月22日自行采用该平台，确定内置助手为夏洛克，Minerva的平台本土多agent AI助手，称 MCP服务器允许运营商通过基于直播遥测的自然语言对话进行分类和调试。 根据Bitdeer AI，该软件可以作为一个独立的平台部署，并与NVIDIA DSX OS集成。 其公告将遥测范围列出为横跨NVIDIA GPU,ConnectX SuperNIC，蓝域DPU,NVLink放大开关，Spectrum-X东西和南北以太网，NCCL，调度器和AI工作量。
报告结果
NVIDIA 称平台在关键度量衡上传递值 AI工厂专注于：中断之间的较长时间，通过在事故发生前发现麻烦；更快的修复时间，通过在事故发生时确定问题；更深入的洞察力，通过过滤数据泛滥现代AI 基础设施生成到少数真正重要的警报。 称公司平台将相关问题卷成一个单一的可操作项目，这样AIOps团队就可以从被动的维护转向主动的维护，帮助AI工厂最大限度地提升GPU的运行时间。
Bitdeer AI报告NVIDIA内部AI工厂测试结果：在一例中，Minerva将修复NVMe电断层的平均时间从多日停电减少到6小时，在另一例中，它将81起吵闹事件倒塌为单一警报，节省了54小时的集群故障时间。 Bitdeer AI 称内部测试还显示，通过预测性维护中断之间的平均时间增加，通过自动根因分析修复的平均时间减少，通过高级降噪来提高可操作性的洞察力，以及提高能直接促进符号生产的有效的GPU时数。
早期收养者和OEM伙伴
IREN是NVIDIA合作伙伴之一，利用Minerva优化其AI工厂。 “NVIDIA Minerva给我们的操作团队一个整个网络的视野，GPU和运行在他们身上的工作，所以我们更注重主动的维护，而不是排除故障。”
Bitdeer AI 称现在支持Minerva的AI工厂业务，称此举是其致力于可靠，高产的GPU基础设施的延伸，用于大规模AI培训和推论。 称公司该平台的可观察性，预测性维护以及代理故障排除有助于它在影响工作之前捕捉退化，缩短修复时间，为客户保持更多的GPU生产，并且该软件的构建是为了从小集群向AI工厂规模化，拥有数十万GPU.
勒诺沃称在自己的GTC柏林站点中，Minerva可以帮助团队在环境尺度上管理AI基础设施的可用性和性能，将其与它的勒诺沃混合AI工厂一起列出，并配有NVIDIA解决方案。 Lenovo还正在就主权AI工厂解决方案与IBM和NVIDIA合作，目的是让客户控制自己的业务。
除IREN外，NVIDIA还命名Aolani,Bitdeer AI,IBM,Jio,Lambda,Quake AI和Tesla为计划部署Minerva. 根据NVIDIA,Dell Technologies,Lenovo和Supermicro将作为AI工厂业务组合的一部分提供平台。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
NVIDIA Launches Minerva AIOps Platform for AI Factory Observability
Add Unite.AI to your preferred sources on Google
NVIDIA launched Minerva on October 22, 2026, describing it as an AIOps platform that gives operations teams earlier warning of infrastructure problems across AI factories. Minerva consolidates telemetry from both infrastructure and AI workloads, then applies purpose-built AI models to convert that data into actionable insight, according to the company.
AI factories rank among the most complex operations in the world, NVIDIA says, with huge fleets of GPUs training models and serving AI applications across high-speed networks. A single degraded component at that scale — a GPU running hotter than its peers or an NVMe drive losing performance under load — can slow an entire job, and such warning signs are easy to miss. Minerva targets what NVIDIA calls AI factory yield: the useful work the infrastructure produces, measured through GPU hours, goodput, cluster uptime and token production.
A Dynamic Knowledge Graph of the Factory
Traditional monitoring tools often split the data center into separate views, with networking, compute, site reliability and machine learning teams each seeing a different part of the operation. NVIDIA says Minerva connects those signals across its GPUs, networking, storage, schedulers and workloads.
From those signals, Minerva builds a dynamic knowledge graph that logically maps GPUs, SuperNICs, switches and cables along with the jobs running on them. NVIDIA says this mapping lets operators trace a slow training job back to the component affecting it, even in a factory running hundreds of thousands of GPUs.
According to NVIDIA, the platform analyzes billions of data points per hour with supervised and unsupervised machine learning, learning what normal behavior looks like, flagging anomalies it has not encountered before, predicting failures as they emerge and correlating related events using the factory’s topology. The platform can consolidate a storm of related events into one focused incident carrying a likely root cause, rather than forwarding each individual symptom to an operator.
AI Assistant and Remediation Workflow
Minerva ships with a native AI assistant and can also connect to external agentic AI tools, giving operators a natural-language interface grounded in live telemetry. Operators can investigate cluster health and receive recommended remediation through conversation. Teams may keep a person in the loop to carry out fixes manually, or connect the platform to automation through infrastructure application programming interfaces and external systems.
Bitdeer AI, which announced its own adoption of the platform on October 22, identifies the built-in assistant as Sherlock, Minerva’s platform-native multi-agent AI assistant, and says an MCP server allows operators to triage and debug through natural-language conversations grounded in live telemetry. According to Bitdeer AI, the software can be deployed as a standalone platform and is integrated with NVIDIA DSX OS. Its announcement lists the telemetry scope as spanning NVIDIA GPUs, ConnectX SuperNICs, BlueField DPUs, NVLink scale-up switches, Spectrum-X East-West and North-South Ethernet, NCCL, schedulers and AI workloads.
Reported Results
NVIDIA says the platform delivers value on the key metrics AI factories focus on: longer time between interruptions, by spotting trouble before it happens; faster time to repair, by pinpointing the problem when an incident occurs; and greater insight, by filtering the flood of data modern AI infrastructure generates down to the few alerts that actually matter. The company says the platform rolls related problems up into a single actionable item so AIOps teams can shift from reactive to proactive maintenance, helping AI factories maximize GPU uptime.
Bitdeer AI reported results from NVIDIA’s internal AI factory testing: in one case, Minerva reduced mean time to repair for an NVMe power fault from a multi-day outage to six hours, and in another, it collapsed 81 noisy events into a single alert, saving 54 hours of cluster uptime. Bitdeer AI said the internal testing also showed an increase in mean time between interrupts through predictive maintenance, a decrease in mean time to repair through automated root-cause analysis, an increase in actionable insights through advanced noise reduction, and an increase in effective GPU hours that directly boosts token production.
Early Adopters and OEM Partners
IREN is among the NVIDIA partners using Minerva to optimize its AI factories. “NVIDIA Minerva gives our operations team one view across the network, the GPUs and the jobs running on them, so we focus more on proactive maintenance rather than troubleshooting,” Denis Skrinnikoff, chief technology officer of IREN, said in the launch announcement.
Bitdeer AI said it now supports Minerva in its AI factory operations, describing the move as an extension of its commitment to reliable, high-yield GPU infrastructure for large-scale AI training and inference. The company said the platform’s observability, predictive maintenance and agentic troubleshooting help it catch degradations before they affect jobs, cut repair time and keep more GPUs productive for customers, and that the software is built to scale from small clusters to AI factories with hundreds of thousands of GPUs.
Lenovo said in its own GTC Berlin post that Minerva can help teams manage AI infrastructure availability and performance as environments scale, listing it alongside its Lenovo Hybrid AI Factory with NVIDIA solutions. Lenovo also said it is collaborating with IBM and NVIDIA on Sovereign AI Factory solutions intended to give customers control over their own operations.
Alongside IREN, NVIDIA named Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla as planning to deploy Minerva. Dell Technologies, Lenovo and Supermicro will offer the platform as part of their AI factory operations portfolios, according to NVIDIA.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.