<?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>Optimization on Duo&#39;s Tech Blog</title>
    <link>https://duoan.github.io/tags/optimization/</link>
    <description>Recent content in Optimization on Duo&#39;s Tech Blog</description>
    <image>
      <title>Duo&#39;s Tech Blog</title>
      <url>https://duoan.github.io/images/papermod-cover.png</url>
      <link>https://duoan.github.io/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.153.1</generator>
    <language>en-us</language>
    <lastBuildDate>Sun, 12 Jul 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://duoan.github.io/tags/optimization/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>The GPU Optimization Playbook: Architecture, Memory, and Balance</title>
      <link>https://duoan.github.io/posts/gpu-optimization-playbook/</link>
      <pubDate>Sun, 12 Jul 2026 00:00:00 +0000</pubDate>
      <guid>https://duoan.github.io/posts/gpu-optimization-playbook/</guid>
      <description>&lt;h1 id=&#34;the-gpu-optimization-playbook-architecture-memory-and-balance&#34;&gt;The GPU Optimization Playbook: Architecture, Memory, and Balance&lt;/h1&gt;
&lt;p&gt;Most &amp;ldquo;GPU optimization&amp;rdquo; advice is a bag of tricks: coalesce here, unroll there, add &lt;code&gt;__restrict__&lt;/code&gt; and pray. Tricks are the &lt;em&gt;output&lt;/em&gt; of optimization, not the method. The method is smaller and more durable: understand the machine, find the resource that is actually saturated, and rebalance work toward the resources that are idle. &lt;strong&gt;Almost every GPU kernel is limited by data movement, not arithmetic.&lt;/strong&gt; Once you internalize that, the whole catalog of techniques collapses into three questions.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
