<?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>ML Networks on punt2cpu</title>
    <link>https://punt2cpu.net/categories/ml-networks/</link>
    <description>Recent content in ML Networks on punt2cpu</description>
    <generator>Hugo</generator>
    <language>en</language>
    <lastBuildDate>Tue, 13 Oct 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://punt2cpu.net/categories/ml-networks/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>ML networking snapshot 202610</title>
      <link>https://punt2cpu.net/2026/10/13/ml-networking-snapshot-202610/</link>
      <pubDate>Tue, 13 Oct 2026 00:00:00 +0000</pubDate>
      <guid>https://punt2cpu.net/2026/10/13/ml-networking-snapshot-202610/</guid>
      <description>&lt;p&gt;My notes on where AI/ML networking stands in October 2026. This area moves fast and there is a lot of active research going on, so parts of this will age quickly. Naming is NVIDIA&amp;rsquo;s (NVLink, NCCL, PXN), AMD and Intel have their own terms for similar things.&lt;/p&gt;
&lt;h2 id=&#34;traffic&#34;&gt;Traffic&lt;/h2&gt;
&lt;p&gt;For comparison, a normal DC Clos carries lots of small TCP flows. ECMP spreads them fine, TCP recovers from drops, 3:1 oversubscription is ok.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
