1<!DOCTYPE html> 2<html lang="en-us"> 3 <head> 4 <meta charset="utf-8"> 5 <meta name="viewport" content="width=device-width, initial-scale=1"> 6 <title> | Zhengrong Wang</title> 7 <link rel="stylesheet" href="/css/style.css" /> 8 <link rel="stylesheet" href="/css/fonts.css" /> 9 10 </head> 11 12 <body> 13 <nav> 14 <ul class="menu"> 15 16 <li><a href="/">Home</a></li> 17 18 <li><a href="/publications/">Publications</a></li> 19 20 <li><a href="/teaching/">Teaching</a></li> 21 22 <li><a href="/posts/">Blog</a></li> 23 24 <li><a href="/random/">Random</a></li> 25 26 <li><a href="/recruiting/">Recruiting</a></li> 27 28 </ul> 29 <hr/> 30 </nav> 31 32<div class="article-meta"> 33<h1><span class="title"></span></h1> 34 35 36</div> 37 38<main> 39<h1 id="interesting-stuff-on-random-topic">Interesting Stuff on Random Topic</h1> 40<h4 id="gpu-related">GPU related</h4> 41<ul> 42<li>2026.8.29 <a href="https://github.com/kacper-daftcode/vLLM-Moet">vLLM-Moet</a> a <em>crazy</em> MoE engine with 2-bit handwritten SASS code (including a reverse-engineer SASS assembler) and SSD MoE offloading. Can serve GLM-5.2 and DSV4-flash on single consumer GPU card (e.g. RTX5090).</li> 43<li>2026.8.4 <a href="https://www.broune.com/blog/designing-ai-chip-hardware-and-software">Designing AI Chip Hardware and Software</a> by Bjarke Hammersholt Roune, a detailed writeup from someone who worked on TPU and GPU design.</li> 44<li>2025.9.4 <a href="https://zhuanlan.zhihu.com/p/1941306442683515068">A Chinese post</a> from <code>melonedo</code> helpful to understanding swizzling</li> 45<li>2025.9.3 <a href="https://github.com/seanzw/random/tree/main/tma_bw">Scripts</a> I used to measure GPU L2 bandwidth.</li> 46<li>2025.7.3 <a href="https://github.com/NVIDIA/cutlass/blob/main/media/docs/cpp/cute/01_layout.md">Cute documentation</a> on layout algebra.</li> 47</ul> 48<h4 id="other">Other</h4> 49<ul> 50<li>Some <a href="/projects">random projects</a> I did for fun during my undergraduate.</li> 51</ul> 52 53</main> 54 55 <footer> 56 <link rel="stylesheet" href="//cdn.jsdelivr.net/npm/katex/dist/katex.min.css">
57<script src="//cdn.jsdelivr.net/combine/npm/katex/dist/katex.min.js,npm/katex/dist/contrib/auto-render.min.js,npm/@xiee/utils/js/render-katex.js" defer></script>
57 58
59<script src="//cdn.jsdelivr.net/npm/@xiee/utils/js/center-img.min.js" defer></script>
59 60 61<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.11.1/styles/default.min.css">
62<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.11.1/highlight.min.js"></script>
62 63 64
65<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.11.1/languages/go.min.js"></script>
65 66
67<script>hljs.highlightAll();</script>
vendor: 72 bytes, lines 67-70
67 68 69 70 <script async src="https://www.googletagmanager.com/gtag/js?id=
70G-CTVHZGKZQ0
vendor: 18 bytes, lines 70-71
70"></script> 71
71<script> 72 var doNotTrack = false; 73 if ( true ) { 74 var dnt = (navigator.doNotTrack || window.doNotTrack || navigator.msDoNotTrack); 75 var doNotTrack = (dnt == "1" || dnt == "yes"); 76 } 77 if (!doNotTrack) { 78
vendor: 157 bytes, lines 78-81
78window.dataLayer = window.dataLayer || []; 79 function gtag(){dataLayer.push(arguments);} 80 gtag('js', new Date()); 81 gtag('config', '
81G-CTVHZGKZQ0
vendor: 12 bytes, lines 81-82
81'); 82
82} 83 </script>
83 84 85 <hr/> 86 © <a href="https://seanzw.github.io">Zhengrong Wang</a> 2026 | <a href="https://github.com/seanzw">Github</a> | Enjoy it! 87 88 </footer> 89 </body> 90</html> 91
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.