<?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[Modified NVIDIA GPUs]]></title><description><![CDATA[<p dir="auto">Some gpu "labs" are modifying NVIDIA GPUs to get way more <strong>VRAM</strong> on them.</p>
<p dir="auto">Probably still one of the best options for the money, but fairly risky.<br />
<strong>NOTE: AFAIK the 5090 96GB are FAKE, do NOT buy them. Beware of scammers!</strong></p>
<p dir="auto">You can find these GPUs easily on eBay or Alibaba for solid pricing, the ones I know of are:</p>
<ul>
<li>RTX 2080ti 22GB</li>
<li>RTX 3080 20GB</li>
<li>RTX 4080 32GB</li>
<li>RTX 4090 48GB</li>
</ul>
<p dir="auto">I myself own 2x RTX 3080 20GB and am very impressed with the results, they are real, and do exist, but don't have working ReBAR support (although a fix for this is <a href="https://airigs.net/topic/7/rebar-alternatives-for-modified-gpus">discussed here</a>, this majorly affects tensor parallel setups (multi-gpu), I highly recommend this tweak).</p>
<p dir="auto">The performance I get seems right, about 80tk/s under heavy context load on Qwen 3.8 27B using <a href="https://huggingface.co/Minachist/Qwen3.8-27B-INT6-Flat-6.6bpw-AutoRound" rel="nofollow ugc">this quant</a> with 256k ctx and vision enabled (with MTP).</p>
<div class="airigs-disclosure airigs-disclosure--no" data-ai-disclosure="no"><i class="fa fa-user" aria-hidden="true"></i><span class="airigs-disclosure__label">No AI used</span></div>]]></description><link>https://airigs.net/topic/12/modified-nvidia-gpus</link><generator>RSS for Node</generator><lastBuildDate>Mon, 02 Nov 2026 07:04:56 GMT</lastBuildDate><atom:link href="https://airigs.net/topic/12.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 30 Sep 2026 16:13:05 GMT</pubDate><ttl>60</ttl></channel></rss>