<?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>Systems on mc · notes</title><link>https://hk.crepuscule.uk/tags/systems/</link><description>Recent content in Systems on mc · notes</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 08 Feb 2026 15:20:00 +0800</lastBuildDate><atom:link href="https://hk.crepuscule.uk/tags/systems/index.xml" rel="self" type="application/rss+xml"/><item><title>KV cache, and why LLM inference is memory-bound</title><link>https://hk.crepuscule.uk/posts/kv-cache/</link><pubDate>Sun, 08 Feb 2026 15:20:00 +0800</pubDate><guid>https://hk.crepuscule.uk/posts/kv-cache/</guid><description>The cache that makes autoregressive decoding fast also makes it the thing that runs out of memory first.</description></item><item><title>Notes on cgroups v2: memory limits that actually hold</title><link>https://hk.crepuscule.uk/posts/cgroups-v2/</link><pubDate>Sun, 09 Mar 2025 21:10:00 +0800</pubDate><guid>https://hk.crepuscule.uk/posts/cgroups-v2/</guid><description>Why my OOM kills moved around after switching to the unified hierarchy, and the three knobs that matter.</description></item></channel></rss>