<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Num_workers on Programmer.ie: Modern AI programming</title>
    <link>http://programmer.ie/tags/num_workers/</link>
    <description>Recent content in Num_workers on Programmer.ie: Modern AI programming</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Sat, 08 Aug 2026 13:21:00 +0100</lastBuildDate>
    <atom:link href="http://programmer.ie/tags/num_workers/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>PyTorch DataLoader Performance: num_workers, pin_memory, Prefetching and Why Your GPU Is Waiting</title>
      <link>http://programmer.ie/post/pytorch-zero-to-hero-05/</link>
      <pubDate>Sat, 08 Aug 2026 13:21:00 +0100</pubDate>
      <guid>http://programmer.ie/post/pytorch-zero-to-hero-05/</guid>
      <description>&lt;h2 id=&#34;pytorch-zero-to-hero--step-05&#34;&gt;PyTorch: Zero to Hero — Step 05&lt;/h2&gt;&#xA;&lt;p&gt;A fast model with a slow input pipeline is still a slow training system.&lt;/p&gt;&#xA;&lt;p&gt;One of the most common PyTorch performance failures looks like this:&lt;/p&gt;&#xA;&lt;div class=&#34;highlight&#34;&gt;&lt;pre tabindex=&#34;0&#34; style=&#34;color:#f8f8f2;background-color:#272822;-moz-tab-size:4;-o-tab-size:4;tab-size:4;&#34;&gt;&lt;code class=&#34;language-text&#34; data-lang=&#34;text&#34;&gt;&lt;span style=&#34;display:flex;&#34;&gt;&lt;span&gt;GPU utilization: 20% → 95% → 10% → 90% → 15%&#xA;&lt;/span&gt;&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;&lt;/div&gt;&lt;p&gt;The model is not necessarily slow.&lt;/p&gt;&#xA;&lt;p&gt;The GPU may simply be waiting for the next batch.&lt;/p&gt;&#xA;&lt;p&gt;This article is about finding out &lt;strong&gt;where the wait is happening&lt;/strong&gt;.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
