<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Lizard inference engineering: Use UMA without a fake copy]]></title><description><![CDATA[<p dir="auto"><strong>Inference Engineering · Day 6 · Evening</strong></p>
<p dir="auto"><img src="https://lizard-llm.qendryx.com/diagrams/lizard-native-stack.png" alt="lizard-native device stack — lizard-llm.qendryx.com" class=" img-fluid img-markdown" /></p>
<p dir="auto">An integrated GPU is not a small discrete GPU.</p>
<p dir="auto">On unified-memory hardware, CPU and GPU already access the same physical memory. Treating that system like a discrete card can add unnecessary staging and copies.</p>
<p dir="auto">lizard-native includes a UMA zero-copy path for supported integrated graphics. The design goal is simple: shared weights should stay shared, while the Direct3D 12 execution path remains explicit.</p>
<p dir="auto">Hardware-aware inference starts by respecting the memory architecture you actually have.</p>
<p dir="auto"><strong>Engineering fact:</strong> lizard-native includes a zero-copy path for unified-memory integrated GPUs so shared memory is treated as shared rather than copied as if discrete.</p>
<p dir="auto"><a href="https://lizard-llm.qendryx.com/docs.html#lizard-native" rel="nofollow ugc">Read the relevant Lizard page</a></p>
<p dir="auto">#LizardNative #IntegratedGPU #D3D12 #ZeroCopy #LocalAI</p>
<p dir="auto">&lt;!-- lizard-marketing-slot:day-06-pm --&gt;</p>
]]></description><link>https://community.lizard-llm.qendryx.com/topic/51/lizard-inference-engineering-use-uma-without-a-fake-copy</link><generator>RSS for Node</generator><lastBuildDate>Mon, 24 Aug 2026 01:37:33 GMT</lastBuildDate><atom:link href="https://community.lizard-llm.qendryx.com/topic/51.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 29 Jul 2026 11:00:12 GMT</pubDate><ttl>60</ttl></channel></rss>