摘要：佩小姐的奇幻城堡...

失业生

China’s ‘Xuelong 2’ begins first ice station survey in central Arctic Ocean_我的网站

刑事侦缉档案4

A |

Chinese research icebreaker Xuelong 2 arrives at the domestic base dock of China's polar expeditions in Waigaoqiao, Shanghai, on May 18, 2026. The vessel successfully completed all tasks of China's 42nd Antarctic expedition mission before returning to Shanghai. Photo: VCGChina's research icebreaker Xuelong 2 began its first ice station survey of its current voyage near 84 degrees north latitude in the central Arctic Ocean on Sunday, marking the start of an approximately half-month comprehensive survey of the region.
After completing the first stage of comprehensive ocean station surveys on Friday, Xuelong 2 sailed toward higher latitudes and selected a large, relatively flat piece of sea ice about 1.5 meters thick in a densely packed ice zone for its first ice station.
According to Lin Long, captain of the Xuelong 2 sea ice team, the ice station survey conducted by the vessel during this voyage will include six short-term ice stations and one long-term ice station, the Xinhua News Agency reported Sunday.
More than 20 types of scientific operations will be carried out, including the deployment of expendable ice-based buoys, snow and ice-core sampling, under-ice water sampling, sea-ice remote sensing, atmospheric profiling and under-ice oceanographic observations.
Lan Musheng, assistant to the chief scientist of the expedition, said the researchers will comprehensively monitor changes in parameters related to the atmosphere, sea ice, upper ocean and ecological processes beneath the ice as the sea ice melts.
The observations will provide scientific data for studying the rapid changes in Arctic sea ice, the mechanisms behind those changes and the evolution of Arctic ecosystems.
The ice station work is part of China's 16th Arctic Ocean scientific expedition, which was organized by the Ministry of Natural Resources. The expedition set sail from Dalian, Northeast China's Liaoning Province on July 3.
The mission is being carried out by four vessels - Xuelong, Xuelong 2, Jidi and Tansuo-3 - and is expected to be completed in early October.
The expedition is focusing on comprehensive surveys of sea ice, hydrology, biology, ecology and the atmospheric environment in key areas of the Arctic Ocean.
The two icebreakers Xuelong and Xuelong 2 will work together on the ice station surveys. Xuelong completed surveys at 12 ice stations on Friday.
Lin said the ice station locations planned for Xuelong 2 will work in coordination with those of Xuelong to form an observation array.
The arrangement will allow researchers to more deeply investigate the interactions among the atmosphere, sea ice and ocean in the central Arctic Ocean amid rapid environmental changes, as well as their coupling effects with the ecosystem.
The expedition is being conducted against a backdrop of rapid changes in the Arctic environment.
When the mission was launched in July, expedition leader Wang Jinhui said that global warming was driving rapid and profound changes in the Arctic natural environment, with trends such as sea-ice melting intensifying.
The expedition is therefore designed not only to monitor environmental changes but also to investigate frontier scientific questions, including the dynamic evolution of the oceanic crust.
Working on sea ice also presents significant safety challenges. Low temperatures and complex ice-surface conditions can pose risks to researchers, while polar bears are considered the biggest safety risk during ice station operations.
The expedition team has therefore formulated specific bear-prevention and emergency response plans and equipped personnel with various types of bear-deterrence equipment and emergency supplies to ensure their safety.
。

B | K图 NVDA_0 　　英伟达硬件演进正加速驶向下一站，为智能体（Agent）打造新一代算力平台。　　当地时间8月24日，英伟达宣布，推理加速器Groq 3 LPX机架已进入全面量产阶段。英伟达高级总监Dion Harris表示，Groq机架将与Vera中央处理器和Rubin图形处理器一同部署在新型云服务商Nebius的数据中心，并将于今年晚些时候上线。　　Groq 3 LPX机架是英伟达专为Vera Rubin平台设计的推理加速扩展组件，需与 Vera Rubin NVL72 GPU主机架搭配部署——NVL72负责更广泛的训练、推理以及上下文处理，LPX专门加速输出AI生成结果（Token），主打超低延迟Token生成，两者异构协同，可大幅提升万亿参数、超长上下文智能体推理的每瓦吞吐与交互响应速度。　　英伟达此前将Vera Rubin平台定义为专为智能体AI与科学计算时代设计的机架级超级计算平台。

C | 英伟达此次将Groq 3 LPX正式推向全面生产，核心目的并不是用专用推理芯片取代GPU，而是为Vera Rubin补充传统GPU架构在低延迟Token生成方面的能力。　　相比Blackwell，Vera Rubin在智能体工作上的效率得到了极大的提升，英伟达释出多个最新测试数据（本次测试使用的基准为SemiAnalysis AgentX，该基准针对Kimi K3、MiniMax M3、GLM5.3、Qwen3.5、DeepSeek V4 Pro等各类模型）—— 　　英伟达称，在Artificial Analysis测试中，搭载Gemma 4 31B模型、100K Token上下文时，Groq 3 LPX实现每秒3400个输出Token，为该模型创下最高纪录；针对智能体编程等低延迟工作负载，其响应速度最高达到最近竞争平台的4倍。英伟达称，这种速度可以让智能体更快完成代码编写、测试、工具调用和结果验证等连续任务。

D | 　　与此同时，英伟达公布Vera Rubin NVL72在真实智能体工作负载下的新测试结果：基于SemiAnalysis的AgentX工作负载、采用DeepSeek V4 Pro模型，Vera Rubin每兆瓦（MW）吞吐量最高达到上一代GB300 NVL72的30倍，每Token成本最高降低35倍。对于运营商而言，这意味着在固定的电力和基础设施预算下，可以支持更高的交互式代理容量，或者以更低的运营成本提供相同的容量。　　英伟达强调，智能体任务不同于传统聊天对话，任务上下文会随着数百个步骤不断累积，甚至达到数十万Token。英伟达援引OpenRouter数据称，智能体AI工作负载消耗的Token数量可以达到简单聊天请求的15倍。原因在于，一个智能体可能先查询数据库，再搜索新闻和文件，调用子智能体进行分析，运行代码并反复验证，整个过程中上下文不断累积。

E | 　　近期腾讯、阿里、DeepSeek等头部厂商相继推进AI智能体产品迭代与生态布局，AI应用端发展聚焦终端场景落地与生态壁垒搭建。天风国际证券最新研报称，当前Harness加速开放迭代，前沿AI竞争正向Agent执行框架与商业化兑现深化。　　英伟达正紧随这一趋势调整硬件优化方向，当AI迈入智能体时代，AI基础设施的评价体系或将随之变化，基础设施服务商的硬件迭代重心将从追求训练更大模型，转向单位能耗有效Token产出、单位成本任务处理能力，以及智能体复杂任务的执行时延。（文章来源：财联社）。