流星蝴蝶剑

Russian warplane crash kills 13_我的网站

苍天有泪

A | IT之家 7 月 22 日消息，英伟达昨日（7 月 21 日）发布博文，从技术角度详细介绍了 Vera CPU，专为智能体 AI（Agentic AI）场景设计。 Yeysk(Russia), Oct 18 (UNI) At least 13 people, including three children, have so far been killed after a Russian fighter-bomber plane crashed into an apartment block in the southern Russian town of Yeysk, officials said Tuesday.

The two pilots of the Su-34 jet, on a training flight, ejected before it hit the building and caused a huge inferno, the BBC reported.

Rescuers revised the death toll after finishing their search of the rubble, Russia's emergencies ministry said.

The officials said 68 were rescued from the blaze in the nine-storey block. Nineteen injured people are being treated following the Monday evening crash.

"According to a report from the pilots, who jettisoned from the plane, the reason for the crash was a fire in one of the engines during take off," the Russian defence ministry said.

"At the point where the Su-34 came down, in the courtyard of a residential block, the plane's fuel supply caught fire," it added.

Footage has emerged apparently showing local residents trying to help one of the pilots, who can be seen lying on the ground with a parachute behind him.

The port town of Yeysk lies near the eastern Ukrainian war zone, across the Sea of Azov from the devastated city of Mariupol. It has been used as a major training ground for Russia's naval aviation.

Pupils from a nearby secondary school were among more than 360 people evacuated from the crash site, Russian media reported.

A local correspondent in Yeysk told Russia's state-run TV channel Rossiya 24 that two of the apartment blocks caught fire, the BBC report said.

The Kremlin has ordered national and regional authorities to provide "all necessary assistance" to victims of the fire.

Russia's Investigative Committee said it had opened a criminal case and sent investigators to the scene.

UNI ING。

B | 英伟达表示智能体 AI 将更多的关键执行路径转移到了 CPU 上。智能体（Agents）在沙箱中运行代码、调用工具、检索上下文、与数据库交互以及分析结果，然后再将信息返回给模型。 IT之家援引博文介绍，智能体 AI 工作负载主要取决于以下几点：即使插槽完全负载，也需要强大的单线程性能。

C | 每个核心都有足够的内存带宽，以持续为许多活跃的执行上下文提供数据。并发下的延迟具有可预测性，以便智能体能够稳定完成预定步骤。高效处理不规则控制流、长依赖链和指针密集型数据结构。英伟达在博文中强调，Vera CPU 围绕智能体 AI 负载设计，重点提升满负载条件下的单线程表现，其设计核心是 Olympus 核心。英伟达强调 Olympus 核心采用前端、核心中段、执行引擎和缓存子系统四部分设计。前端集成 10 宽解码引擎，并支持每周期最多处理 2 个已采用分支。其分支预测子系统包含神经分支预测器，用于提升复杂分支模式下的预测准确性。该公司表示，Vera CPU 共配备 88 个 Olympus 核心和 176 个 SMT 线程，目标是在单线程高负载与更高线程密度两类场景间切换，并降低线程间资源争用。在片上互连与缓存部分，Vera CPU 搭载 NVIDIA 可扩展相干互连，即 NVIDIA Scalable Coherency Fabric（SCF）。