<?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[7900xt或者7900xt、w7800的选择是avalon]]></title><description><![CDATA[<p dir="auto">7900xt或者7900xt、w7800的选择是avalon，同样时间发布的版本情况下，比strata 快40%以上，而且是专门支持rdna 3，特别优化过的。测试的话，各位看官自己测吧。</p>
]]></description><link>https://lcz.me/topic/2842</link><generator>RSS for Node</generator><lastBuildDate>Mon, 09 Nov 2026 11:54:30 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2842.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 04 Nov 2026 17:37:45 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 07:16:20 GMT]]></title><description><![CDATA[<p dir="auto">用现有硬件折腾出最高效果。才是折腾的意义！</p>
]]></description><link>https://lcz.me/post/31279</link><guid isPermaLink="true">https://lcz.me/post/31279</guid><dc:creator><![CDATA[williamlouis]]></dc:creator><pubDate>Thu, 05 Nov 2026 07:16:20 GMT</pubDate></item><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 07:07:17 GMT]]></title><description><![CDATA[<p dir="auto">从两者实现的原理上看，strata串行，avalon并行。如果想多进程调用，avalon会有很大优势。</p>
]]></description><link>https://lcz.me/post/31277</link><guid isPermaLink="true">https://lcz.me/post/31277</guid><dc:creator><![CDATA[vanvj]]></dc:creator><pubDate>Thu, 05 Nov 2026 07:07:17 GMT</pubDate></item><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 06:33:46 GMT]]></title><description><![CDATA[<p dir="auto">两边版本都在不停迭代，同一时间，最好是同一个配置，才能所谓看出高低。定制的引擎终究要好些。</p>
]]></description><link>https://lcz.me/post/31265</link><guid isPermaLink="true">https://lcz.me/post/31265</guid><dc:creator><![CDATA[vanvj]]></dc:creator><pubDate>Thu, 05 Nov 2026 06:33:46 GMT</pubDate></item><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 03:09:49 GMT]]></title><description><![CDATA[<p dir="auto">定制框架越来越多了，我看野火的数据挺好的</p>
]]></description><link>https://lcz.me/post/31178</link><guid isPermaLink="true">https://lcz.me/post/31178</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:09:49 GMT</pubDate></item><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 03:06:54 GMT]]></title><description><![CDATA[<p dir="auto">我就是7900xtx24g，这个要怎么测试</p>
]]></description><link>https://lcz.me/post/31174</link><guid isPermaLink="true">https://lcz.me/post/31174</guid><dc:creator><![CDATA[ma yi]]></dc:creator><pubDate>Thu, 05 Nov 2026 03:06:54 GMT</pubDate></item><item><title><![CDATA[Reply to 7900xt或者7900xt、w7800的选择是avalon on Thu, 05 Nov 2026 00:28:47 GMT]]></title><description><![CDATA[<p dir="auto"><a href="https://github.com/Niko1221/Strata/issues/5113?utm_source=chatgpt.com" rel="nofollow ugc">https://github.com/Niko1221/Strata/issues/5113?utm_source=chatgpt.com</a><br />
Avalon 的优势主要在 Decode，而不是 Prefill，综合对比还是Strata RDNA3 Fork更佳吧：</p>
<h2>5. Numbers / 数据</h2>
<h3>5.1 Prefill (t/s, median) / Prefill（t/s，中位数）</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>Case / 场景</th>
<th>fork 0.1.49.1</th>
<th>official 0.1.55</th>
<th>0.1.55 W8</th>
<th>Avalon</th>
</tr>
</thead>
<tbody>
<tr>
<td>32K</td>
<td><strong>3,484</strong></td>
<td>3,198</td>
<td>3,216</td>
<td>2,044</td>
</tr>
<tr>
<td>64K</td>
<td><strong>3,514</strong></td>
<td>3,261</td>
<td>3,267</td>
<td>1,981</td>
</tr>
<tr>
<td>131K</td>
<td><strong>3,332</strong></td>
<td>3,220</td>
<td>3,221</td>
<td>1,861</td>
</tr>
<tr>
<td>250K</td>
<td><strong>3,174</strong></td>
<td>3,060</td>
<td>3,060</td>
<td>1,675</td>
</tr>
</tbody>
</table>
<h3>5.2 Generation (t/s, median) / 生成（t/s，中位数）</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>Case / 场景</th>
<th>fork 0.1.49.1</th>
<th>official 0.1.55</th>
<th>0.1.55 W8</th>
<th>Avalon</th>
</tr>
</thead>
<tbody>
<tr>
<td>short / 短提示</td>
<td>103.2</td>
<td>82.8</td>
<td>84.9</td>
<td><strong>117.5</strong></td>
</tr>
<tr>
<td>32K</td>
<td>88.5</td>
<td>72.7</td>
<td>73.6</td>
<td><strong>105.9</strong></td>
</tr>
<tr>
<td>64K</td>
<td>81.3</td>
<td>72.0</td>
<td>72.4</td>
<td><strong>102.2</strong></td>
</tr>
<tr>
<td>131K</td>
<td>84.3</td>
<td>70.5</td>
<td>69.7</td>
<td><strong>96.5</strong></td>
</tr>
<tr>
<td>250K</td>
<td>82.7</td>
<td>68.4</td>
<td>64.3</td>
<td><strong>95.7</strong></td>
</tr>
</tbody>
</table>
<h3>5.3 MTP acceptance / MTP 接受率</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>Case / 场景</th>
<th>fork</th>
<th>official 0.1.55</th>
<th>0.1.55 W8</th>
<th>Avalon</th>
</tr>
</thead>
<tbody>
<tr>
<td>32K</td>
<td>81%</td>
<td>77%</td>
<td>77%</td>
<td>66%</td>
</tr>
<tr>
<td>64K</td>
<td>81%</td>
<td>81%</td>
<td>81%</td>
<td>61%</td>
</tr>
<tr>
<td>131K</td>
<td>84%</td>
<td>79%</td>
<td>81%</td>
<td>57%</td>
</tr>
<tr>
<td>250K</td>
<td>86%</td>
<td>83%</td>
<td>86%</td>
<td>63%</td>
</tr>
</tbody>
</table>
]]></description><link>https://lcz.me/post/31120</link><guid isPermaLink="true">https://lcz.me/post/31120</guid><dc:creator><![CDATA[李野火]]></dc:creator><pubDate>Thu, 05 Nov 2026 00:28:47 GMT</pubDate></item></channel></rss>