<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Occupancy on icyveins7's blog</title><link>https://icyveins7.github.io/tags/occupancy/</link><description>Recent content in Occupancy on icyveins7's blog</description><generator>Hugo</generator><language>en-us</language><copyright>This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.</copyright><lastBuildDate>Wed, 12 Aug 2026 20:00:00 +0800</lastBuildDate><atom:link href="https://icyveins7.github.io/tags/occupancy/index.xml" rel="self" type="application/rss+xml"/><item><title>Occupancy-maxxing is just Starcraft</title><link>https://icyveins7.github.io/posts/2026/08/occupancy-maxxing-is-just-starcraft/</link><pubDate>Wed, 12 Aug 2026 20:00:00 +0800</pubDate><guid>https://icyveins7.github.io/posts/2026/08/occupancy-maxxing-is-just-starcraft/</guid><description>&lt;blockquote&gt;&#10;&lt;p&gt;If you can manage minerals and gas, you can manage CUDA threads and registers.&lt;/p&gt;&#10;&lt;/blockquote&gt;&#10;&lt;blockquote&gt;&#10;&lt;p&gt;CUDA is just resource management.&lt;/p&gt;&#10;&lt;/blockquote&gt;&#10;&lt;p&gt;Okay, of course there&amp;rsquo;s a lot more to squeezing CUDA kernel performance than &lt;em&gt;just resource management&lt;/em&gt;, but occupancy is very often the simplest thing to strive for, once you know how. And usually, if you&amp;rsquo;ve hit 100% occupancy, then unless you&amp;rsquo;ve done something heinous in your code, further optimizations are unlikely to budge your performance by large factors (aside from an entire algorithmic shift, but we won&amp;rsquo;t discuss that here).&lt;/p&gt;</description></item></channel></rss>