<?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>Homelab on debene.dev</title>
    <link>https://debene.dev/categories/homelab/</link>
    <description>Recent content in Homelab on debene.dev</description>
    <image>
      <title>debene.dev</title>
      <url>https://debene.dev/images/og-default.png</url>
      <link>https://debene.dev/images/og-default.png</link>
    </image>
    <generator>Hugo -- 0.152.2</generator>
    <language>en-us</language>
    <lastBuildDate>Thu, 14 May 2026 14:00:00 -0500</lastBuildDate>
    <atom:link href="https://debene.dev/categories/homelab/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Running Modern LLMs on a 2016 IBM POWER8 in 2026</title>
      <link>https://debene.dev/posts/power8-llm-2026/</link>
      <pubDate>Thu, 14 May 2026 14:00:00 -0500</pubDate>
      <guid>https://debene.dev/posts/power8-llm-2026/</guid>
      <description>&lt;h2 id=&#34;what-are-we-even-doing-here&#34;&gt;What Are We Even Doing Here?&lt;/h2&gt;
&lt;p&gt;It&amp;rsquo;s 2026. Most people run LLMs on NVIDIA H100s, AMD MI300X, or at least a decent gaming GPU. I&amp;rsquo;m running them on a 2016 IBM POWER8 server with 160 hardware threads and zero CUDA cores.&lt;/p&gt;
&lt;p&gt;Why? Because I can. And because nobody else has published POWER8 LLM benchmarks in 2026. And because alternative architectures deserve love too.&lt;/p&gt;
&lt;p&gt;This post covers:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Building llama.cpp on ppc64le with GCC 16&lt;/li&gt;
&lt;li&gt;Running Qwen 2.5 7B (text + vision) on POWER8&lt;/li&gt;
&lt;li&gt;NUMA tuning discoveries (spoiler: conventional wisdom is wrong)&lt;/li&gt;
&lt;li&gt;Multimodal inference (yes, vision models work too)&lt;/li&gt;
&lt;li&gt;Full reproduceability (Gentoo USE flags, build commands, everything)&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;TL;DR:&lt;/strong&gt; Got 6.81 tokens/s on text generation and fully functional vision inference. POWER8 reads license plates better than some humans.&lt;/p&gt;</description>
    </item>
    <item>
      <title>🌹 ULTRA2: O Amor Que Não Compilava</title>
      <link>https://debene.dev/posts/ultra2-telenovela-pt/</link>
      <pubDate>Sat, 09 May 2026 00:00:00 +0000</pubDate>
      <guid>https://debene.dev/posts/ultra2-telenovela-pt/</guid>
      <description>Uma telenovela brasileira em 8 capítulos sobre como construir um node K8s com NPU, RDMA, e muita treta com Ubuntu 26.04</description>
    </item>
    <item>
      <title>🌹 ULTRA2: The Love That Wouldn&#39;t Compile</title>
      <link>https://debene.dev/posts/ultra2-telenovela-en/</link>
      <pubDate>Sat, 09 May 2026 00:00:00 +0000</pubDate>
      <guid>https://debene.dev/posts/ultra2-telenovela-en/</guid>
      <description>A Brazilian-style soap opera in 8 chapters about building a Kubernetes node with NPU, RDMA, and a lot of drama with Ubuntu 26.04</description>
    </item>
    <item>
      <title>The 285H That Cried Meh: A Six-Hour Geekbench Saga Through Talos, UKI, and BIOS Power Limits</title>
      <link>https://debene.dev/posts/ultra2-diagnostic-saga/</link>
      <pubDate>Fri, 08 May 2026 00:00:00 +0000</pubDate>
      <guid>https://debene.dev/posts/ultra2-diagnostic-saga/</guid>
      <description>How a brand-new Intel Core Ultra 9 285H benched at half its potential, and the rabbit hole that fixed it.</description>
    </item>
    <item>
      <title>Cross-Flashing a Mellanox ConnectX-3 Pro: Or How I Learned to Stop Worrying and Love InfiniBand</title>
      <link>https://debene.dev/posts/mellanox-connectx3-crossflash/</link>
      <pubDate>Wed, 06 May 2026 13:18:00 -0500</pubDate>
      <guid>https://debene.dev/posts/mellanox-connectx3-crossflash/</guid>
      <description>OEM firmware locked your ConnectX card to Ethernet-only? Here&amp;#39;s how to cross-flash it to VPI mode, survive the GUID apocalypse, and get 40 Gbps InfiniBand running in your homelab.</description>
    </item>
    <item>
      <title>The $400 Sovereignty Stack: Why I&#39;m Rebuilding My Homelab Around a Laptop CPU Soldered to a Board from Shenzhen</title>
      <link>https://debene.dev/posts/400-dollar-sovereignty-stack/</link>
      <pubDate>Wed, 06 May 2026 11:45:00 -0500</pubDate>
      <guid>https://debene.dev/posts/400-dollar-sovereignty-stack/</guid>
      <description>Why I bought a laptop CPU soldered to a motherboard from Shenzhen, what I&amp;#39;m actually paying for when I build infrastructure, and why the homelab renaissance isn&amp;#39;t about saving money.</description>
    </item>
  </channel>
</rss>
