<?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>AI芯片 on FisherAI</title>
    <link>https://fisherdaddy.com/tags/ai%E8%8A%AF%E7%89%87/</link>
    <description>Recent content in AI芯片 on FisherAI</description>
    <image>
      <title>FisherAI</title>
      <url>https://fisherdaddy.com/images/papermod-cover.png</url>
      <link>https://fisherdaddy.com/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.125.7</generator>
    <language>en</language>
    <lastBuildDate>Tue, 05 May 2026 10:02:00 +0800</lastBuildDate>
    <atom:link href="https://fisherdaddy.com/tags/ai%E8%8A%AF%E7%89%87/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>为什么 AI 会有 Fast Mode？Reiner Pope 用一块黑板讲透芯片、推理与集群经济学</title>
      <link>https://fisherdaddy.com/posts/reiner-pope-ai-chips-inference-economics/</link>
      <pubDate>Tue, 05 May 2026 10:02:00 +0800</pubDate>
      <guid>https://fisherdaddy.com/posts/reiner-pope-ai-chips-inference-economics/</guid>
      <description>MatX CEO、前 Google TPU 架构师 Reiner Pope 从批大小、KV cache、MoE、scale-up 网络和内存墙出发，解释了大模型推理价格、延迟、长上下文、稀疏模型与训练/推理经济学背后的硬约束。</description>
    </item>
  </channel>
</rss>
