<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>机架级 on AI 实战派 · 从技术到赚钱</title><link>https://guijiagi.com/tags/%E6%9C%BA%E6%9E%B6%E7%BA%A7/</link><description>Recent content in 机架级 on AI 实战派 · 从技术到赚钱</description><generator>Hugo</generator><language>zh-cn</language><copyright>本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权</copyright><lastBuildDate>Thu, 08 Oct 2026 16:00:00 +0800</lastBuildDate><atom:link href="https://guijiagi.com/tags/%E6%9C%BA%E6%9E%B6%E7%BA%A7/index.xml" rel="self" type="application/rss+xml"/><item><title>NVIDIA Dynamo：分离式推理为什么到 2026 才在机架级落地</title><link>https://guijiagi.com/posts/2026-10-08-11-nvidia-dynamo-disaggregation/</link><pubDate>Thu, 08 Oct 2026 16:00:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-08-11-nvidia-dynamo-disaggregation/</guid><description>Dynamo 1.0 四月发布、六月迭代到 v1.2，把 prefill 与 decode 拆到不同资源池。本文解释这个早就被讨论的概念为什么今年才真正可用。</description></item><item><title>Vera CPU 为什么是「给智能体的 CPU」：Agent 负载把 CPU 重新变成瓶颈</title><link>https://guijiagi.com/posts/2026-10-08-04-nvidia-vera-cpu-agentic/</link><pubDate>Thu, 08 Oct 2026 15:18:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-08-04-nvidia-vera-cpu-agentic/</guid><description>NVIDIA Vera CPU 号称比 x86 快 1.8 倍，专为 Agentic AI 与强化学习设计。本文解释为什么 Agent 时代 CPU 密集型工作负载回来了。</description></item><item><title>Vera Rubin NVL72 量产拆解：20.7TB HBM4 为什么比算力更重要</title><link>https://guijiagi.com/posts/2026-10-08-03-vera-rubin-nvl72-hbm4/</link><pubDate>Thu, 08 Oct 2026 15:12:00 +0800</pubDate><guid>https://guijiagi.com/posts/2026-10-08-03-vera-rubin-nvl72-hbm4/</guid><description>Vera Rubin NVL72 每机架 20.7TB HBM4、1.6PB/s 带宽、3600 PFLOPS NVFP4。本文解释 Agent 时代为什么容量墙比算力墙先到。</description></item></channel></rss>