<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Tensor Ray</title><link>https://blog.mbinf.de/</link><description>Recent content on Tensor Ray</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Tue, 30 Jun 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://blog.mbinf.de/index.xml" rel="self" type="application/rss+xml"/><item><title>An agent runs my GPU cluster</title><link>https://blog.mbinf.de/posts/agent-runs-my-gpu-cluster/</link><pubDate>Tue, 30 Jun 2026 00:00:00 +0000</pubDate><guid>https://blog.mbinf.de/posts/agent-runs-my-gpu-cluster/</guid><description>&lt;p&gt;For three weeks an agent has been running my GPU cluster. It scales Blackwell cards up and down, picks nodes, and frees silicon when nobody is rendering.&lt;/p&gt;
&lt;p&gt;Not autonomous magic. It runs the loop I used to babysit by hand. Here is the honest version of what that taught me, with the parts LinkedIn was too short to hold.&lt;/p&gt;
&lt;p&gt;&lt;img src="https://blog.mbinf.de/images/story-a-hero.png" alt="A GPU rack orchestrated by an agent"&gt;&lt;/p&gt;
&lt;h2 id="the-first-lesson-idle-gpus-that-are-still-blocked"&gt;The first lesson: idle GPUs that are still blocked&lt;/h2&gt;
&lt;p&gt;A render UI grabs a GPU the moment its process starts, not when you click &amp;ldquo;render&amp;rdquo;. &lt;code&gt;torch.cuda&lt;/code&gt; takes the device on import. So an interface nobody is using still owns a card all day.&lt;/p&gt;</description></item></channel></rss>