<?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>Gemma3 on Aillex — Do-It-Yourself AI</title>
    <link>https://askaillex.com/tags/gemma3/</link>
    <description>Recent content in Gemma3 on Aillex — Do-It-Yourself AI</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Fri, 11 Sep 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://askaillex.com/tags/gemma3/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Local AI on 8 GB of VRAM: What Actually Fits, and What Actually Runs</title>
      <link>https://askaillex.com/guides/local-ai-8gb-vram/</link>
      <pubDate>Fri, 11 Sep 2026 00:00:00 +0000</pubDate>
      <guid>https://askaillex.com/guides/local-ai-8gb-vram/</guid>
      <description>We took a stock Asus TUF gaming laptop with an RTX 3070 and 8 GB of video memory, and benched eight local models overnight across ten tasks. The result that surprised us: a 30 billion parameter model with two thirds of itself off the card beat a 14 billion parameter model that fit better, by nearly four times. File size does not predict speed. All 103 measured rows are free below, failures included.</description>
    </item>
  </channel>
</rss>
