跳到正文

#部署/工程

今日 33 条
10月2日周五
  1. Hacker News:AI 热帖62

    TSMC 考虑在得州建设数十亿美元园区以扩大 AI 芯片产能

    TSMC 正考虑在美国得州建设新园区,规划包含多座晶圆厂,每座成本至少 200 亿美元,以应对 AI 硬件需求。公司已承诺在亚利桑那州投入 2650 亿美元;得州项目仍处早期阶段,并取决于美国延长先进制造税收抵免。TSMC 同时也与新加坡官员进行过沟通,公司对市场传闻不予置评。

  2. Hacker News:AI 热帖47

    初创公司 Corbenic AI 推出 Galahad,保护 AI 记忆并降低能耗

    比利时初创公司 Corbenic AI 推出测试版工具 Galahad,用于加密 AI 存储记忆、隔离客户信息,并记录 AI 活动以便开发者审查操作和恢复记忆状态。内部测试中,Galahad 通过复用已处理结果,将重复查询的 GPU 能耗降低 92-93%,且无需修改底层模型。该测试版为非商业单 GPU 系统提供 12 个月免费许可,支持 vLLM、SGLang 和 llama.cpp。

  3. 🚨 AI News | TestingCatalog57

    Volantis 宣布完成 8800 万美元 A 轮融资,由 Lachy Groom 和 Abstract 共同领投,John Doerr、Susa Ventures、VXI Capital、Triatomic Capital 参投。

    引用Tapa Ghosh@semiDL

    Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: https://volantissemi.ai/news-insights/our-88m-series-a-demolishing-the-memory-wall-with-photonics-post

  4. Hacker News:AI 热帖49

    Show HN:AilaFlow,面向团队的自托管 AI 工作空间

    AilaFlow 是面向团队的自托管 AI 工作空间,可连接工具、数据与工作流,并通过 AI 聊天和表单执行任务、设置人工审批。管理员可用 AI 或可视化构建器创建流程,配置权限、可用动作和审批点;工作流代码运行在隔离的 Docker 沙箱中。该工具需安装 Docker,可通过 `npm install -g @ailaflow/cli` 和 `ailaflow serve` 启动。

  5. Rohan Paul53

    Volantis 宣布完成 8800 万美元 A 轮融资,累计融资 9700 万美元,投资方包括 Sam Altman、Jeff Dean 和 John Doerr。

    引用Tapa Ghosh@semiDL

    Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: https://volantissemi.ai/news-insights/our-88m-series-a-demolishing-the-memory-wall-with-photonics-post

  6. elvis46

    我认为更快的推理是编码智能体接下来的重大突破之一。 Volantis 正利用光学技术,为每颗芯片提供大得多的内存和高得多的内存带宽。他们的目标是在参数量超过 10T 的模型上,实现每用户每秒最高 10,000 tokens。这太疯狂了! 在那样的速度下,如今需要数小时的编码智能体,几分钟内就能完成。 这绝对是我近期见过更令人兴奋的融资之一。

    引用Tapa Ghosh@semiDL

    Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: https://volantissemi.ai/news-insights/our-88m-series-a-demolishing-the-memory-wall-with-photonics-post

  7. Databricks:Blog(RSS)49

    Lakebase Postgres 基于分支的恢复如何实现大规模快速恢复

    Lakebase Postgres 推出基于分支的恢复机制,将恢复变为在指定时间戳创建分支的元数据操作,即使数据库达 100 TB,恢复时间也可降至秒级。其架构将计算与存储解耦,由 Safekeepers、pageserver 和对象存储分别负责 WAL 持久化、页面重构与长期历史保存,避免传统 PITR 部署实例、拉取快照和重放 WAL 的多小时等待。

  8. AWS Machine Learning Blog46

    在 AWS 上为 Claude Platform 实现多环境访问

    AWS 教程展示如何让 AWS 生产负载、开发者笔记本和外部 CI/CD 三类环境共用 Claude Platform on AWS 订阅,并通过 workspace 隔离生产与开发流量。方案采用专用 AI Services 账户,分别配置跨账户 SigV4、workspace 级 API key 和 OIDC 联邦短期凭证。教程包含 CLI 命令、控制台操作与代码片段。

  9. Hacker News:AI 热帖65

    DeepSeek 与华为合作开发先进芯片软件工具,对标 Nvidia CUDA 生态

    DeepSeek 宣布已为华为技术制造的芯片开发软件工具,并在官方微信公众号称其可作为 CUDA 的潜在替代方案。报道指出,中国发展本土 AI 产业的主要障碍之一是让芯片发挥性能的底层软件,而 Nvidia 的优势不仅在芯片设计,还包括全球约 400 万使用 CUDA 的软件开发者。

10月1日周四
  1. elvis38

    搭建你自己的测试框架吧,各位! 搭建我自己的元测试框架,最大的好处之一就是我能极其顺畅地在不同模型之间切换。 我感觉随着模型发布周期缩短,这一点正变得越来越重要。RSI 只会进一步扩大发布规模。 这是我自去年 11 月以来做过的最正确的决定之一。 切换模型确实可能把东西搞坏,但本不该如此。如果真的搞坏了,说明你的配置并不适应当前这种疯狂的发布节奏。你现在就得解决这个问题。