资讯与观察
AI 行业动态、工程实践、产品方法和社区故事。
OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide
As AI factories grow from megawatts to gigawatts, building them takes more than advanced computing. Open standards are helping the industry scale AI infrastructure faster, more efficiently and more reliably. At the Open…
F1 in Singapore: There's a new fastest car now
Also, F1’s final sprint saw our first real wet racing of the year.
4 分钟 · Ars Technica
Google announces the Fitbit Edge, a middle ground between Pixel Watch 5 and Fitbit Air
The Fitbit Edge offers more than the Air, but it's not a real smartwatch.
4 分钟 · Ars TechnicaLATEST STORIES
最新内容
Amazon has over 1 million robot workers. What’s next for human employees?精选导读
The second-largest corporate employer has bet big on robot workers to scale.
AMD EPYC 8005 SKU List and Value Analysis: Mainstream Sorano’s Big Stride精选导读
Following the recent launch of AMD's EPYC 8005 Sorano server CPUs, we are taking a look at the various SKU options among chips, and what configurations offer the best value for different needs
Review: 4-hour Musk documentary feels like two different films精选导读
Alex Gibney's provocative film is compelling but overlong, and there's not much new here for Ars readers.
Prime Video unveils Blade Runner 2099 full trailer at NYCC精选导读
Also: New trailer for Godzilla Minus Zero, the sequel to Takashi Yamazaki’s kaiju epic Godzilla Minus One.
SPECviewperf 15.0.1 Linux Edition Released: Linux’s Key Graphics Benchmark Adds Arm Support精选导读
SPEC has released version 15.0.1 of SPECviewperf Linux Edition, bringing a slew of upgrades and official Arm support to Linux's key workstation graphics benchmark suite. We are taking a look at those changes, and what p…
Pallas 内核怎么调优？Google 给出的答案是“少信模型，多看寄存器”精选导读
Google 为开源 TPU 性能分析器 XProf 新增内核性能分析套件，可展示 Pallas 自定义内核的周期级细节，此前这些内核在跟踪记录中仅显示为单个不透明块。在 TPU v7（Ironwood）上，XProf 会在运行时采样硬件性能计数器。一个分块矩阵乘法示例中，Google 借此发现内存停顿问题，通过添加三重缓冲将内核执行时间从 125.5µs 缩短至 88µs，降幅约 30%。
消息称英伟达计划以 24GB 显存 GeForce RTX 5080 显卡取代现有 16GB 版本精选导读
英伟达计划停产 16GB 显存的现有 GeForce RTX 5080，转而推出 24GB 显存版本，预计 2027 年第 1 季度面世，AIC 合作伙伴预计近期收到开发资料。新卡预计配备 8 颗 24Gb（3GB）GDDR7 显存，是否提升显存等效速率和整卡功耗尚不清楚。
长鑫技术新突破，融合 4F² 架构等先进技术的 DDR5 RDIMM 产品预计年底亮相精选导读
长鑫存储 4F² DRAM 架构研发取得关键进展，总裁曹堪宇在第四届集成芯片和芯粒大会上宣布，预计年底推出融合 4F² 架构、混合键合等技术的新一代 DDR5 RDIMM 产品。长鑫自成立起布局 4F²，公开专利族超 300 件，2021 年发布大陆首篇相关论文。演讲未提及该产品的规格与具体发布时间。
消息称 Meta 已搁置向 Anthropic 出租 AI 算力的计划精选导读
Meta 已搁置向 Anthropic 出租 AI 算力的计划。Anthropic CEO Dario Amodei 今年早些时候曾致电 Meta 首席人工智能官汪滔，希望租赁 AI 计算芯片，Meta 内部讨论后决定暂不提供。随着 Muse 系列自研模型发布，Meta 自身算力需求迅速增长。