<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>homelab on Purinda Gunasekara</title>
		<link>https://purinda.com/tags/homelab/</link>
		<description>Recent content in homelab on Purinda Gunasekara</description>
		<generator>Hugo</generator>
		<language>en-AU</language>
		
		
		
		
			<lastBuildDate>Thu, 29 Oct 2026 22:45:00 +1100</lastBuildDate>
		
			<atom:link href="https://purinda.com/tags/homelab/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>Capping a Radeon RX 7900 XTX for LLM inference: a step-by-step</title>
				<link>https://purinda.com/posts/underclocking-rx-7900-xtx/</link>
				<pubDate>Thu, 29 Oct 2026 22:45:00 +1100</pubDate>
				<guid>https://purinda.com/posts/underclocking-rx-7900-xtx/</guid>
				<description>A 7900 XTX serving a 27B LLM drew 315 W at full load. The VBIOS floor is 283 W, the firmware refused to pin clocks, and the whole story turned on one property of the workload: the card is memory-bandwidth-bound. 315 W to 283 W, the junction 91 °C to 83 °C, and one token per second off.</description>
			</item>
	</channel>
</rss>
