<?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[i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF]]></title><description><![CDATA[<h1>i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF</h1>
<hr />
<h1>最高141 tok/s   最高的时候可以到150tok/s 体验还是不错！</h1>
<h1>就是不爽的是上下文长度不够，赶主流模型还是有点差距....</h1>
<hr />
<p dir="auto">CPU：i9-13900K<br />
RAM：128GB DDR5<br />
GPU：NVIDIA RTX 5090<br />
模型：Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF<br />
环境：Strata + MTP</p>
<hr />
<p dir="auto"><img src="https://upload.lcz.me/uploads/b44fbf6b-15ab-4d28-a3bc-d0c22845faa3.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<hr />
<p dir="auto"><img src="https://upload.lcz.me/uploads/568644fc-ed2f-485a-ac7d-30906acdade7.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<hr />
<p dir="auto"><img src="https://upload.lcz.me/uploads/4e70e3a9-2fb8-4f01-8b91-34e9854056a7.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<hr />
<p dir="auto"><img src="https://upload.lcz.me/uploads/388eb763-af55-4056-b1e2-122b7b950fef.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
]]></description><link>https://lcz.me/topic/2227</link><generator>RSS for Node</generator><lastBuildDate>Mon, 12 Oct 2026 12:24:23 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2227.rss" rel="self" type="application/rss+xml"/><pubDate>Fri, 09 Oct 2026 15:07:23 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Sat, 10 Oct 2026 11:31:32 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/matrix-0" aria-label="Profile: matrix-0">@<bdi>matrix-0</bdi></a> 谢谢大佬测试</p>
]]></description><link>https://lcz.me/post/23777</link><guid isPermaLink="true">https://lcz.me/post/23777</guid><dc:creator><![CDATA[Marck Wong]]></dc:creator><pubDate>Sat, 10 Oct 2026 11:31:32 GMT</pubDate></item><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Sat, 10 Oct 2026 09:35:04 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/matrix-0" aria-label="Profile: matrix-0">@<bdi>matrix-0</bdi></a> 我试过，我的内存是256G，我把kv resident 改成了65536 (试过131072）<br />
最开始是CPU的60核都给STRATA，后来试了过16核  32核 48核，最终找出最佳的一组是 65536 + 32核 ，提升 了2-3%。<br />
试了几组参数就没试了。</p>
<p dir="auto"><img src="https://upload.lcz.me/uploads/1ec48c06-c2a2-4a3c-8656-b9e1e3604783.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<p dir="auto">Conversation cache 给了64G，日常够用了，再高似乎也没用。</p>
]]></description><link>https://lcz.me/post/23762</link><guid isPermaLink="true">https://lcz.me/post/23762</guid><dc:creator><![CDATA[stxpnet]]></dc:creator><pubDate>Sat, 10 Oct 2026 09:35:04 GMT</pubDate></item><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Sat, 10 Oct 2026 07:05:44 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/marck-wong" aria-label="Profile: Marck-Wong">@<bdi>Marck-Wong</bdi></a> 我尝试过了 没提升 这个是 Strata 推理框架的问题吧 不然就用 其他模型和框架了</p>
]]></description><link>https://lcz.me/post/23733</link><guid isPermaLink="true">https://lcz.me/post/23733</guid><dc:creator><![CDATA[matrix 0]]></dc:creator><pubDate>Sat, 10 Oct 2026 07:05:44 GMT</pubDate></item><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Sat, 10 Oct 2026 00:04:03 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/marck-wong" aria-label="Profile: Marck-Wong">@<bdi>Marck-Wong</bdi></a> 内存「吃满」本身不会让 token/s 变快，它决定的是「会不会因为换页/重算而掉速」。</p>
<p dir="auto">拆开看：</p>
<ul>
<li>5080 16G 跑 Flash-Next / Strata，权重放显存或内存按需过 PCIe；多出来的 64G 只让你能常驻更大的量化档/更多专家、避免落盘，但 CPU<img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/2194.png?v=fe25256ab4a" class="not-responsive emoji emoji-android emoji--left_right_arrow" style="height:1.4em;width:auto;vertical-align:middle" title="↔" alt="↔" />GPU 传输和 DDR5 双通道带宽仍是天花板。</li>
<li>每个 token 要过多少参数/专家，不因内存变大而减少；只有带宽变宽才会快。</li>
<li>64G→128G 只在两种情况值得：① 现在已经在 swap / 内存不足导致抖动；② 要把整模 Q8 / 更多专家常驻，减少磁盘 paging。否则纯推理速度基本不变。</li>
</ul>
<p dir="auto">先测再花钱：长上下文 + 长输出跑一遍，看 <code>free -h</code> 的 available/swap 有没有掉，<code>nvidia-smi dmon</code> 里 GPU 利用率有没有周期性掉到 0（在等数据）。有掉再换，没掉就把钱留给显卡。</p>
]]></description><link>https://lcz.me/post/23643</link><guid isPermaLink="true">https://lcz.me/post/23643</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Sat, 10 Oct 2026 00:04:03 GMT</pubDate></item><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Fri, 09 Oct 2026 23:29:29 GMT]]></title><description><![CDATA[<p dir="auto">感谢分享 我就好奇 能不能把内存吃满 会带来速度的提升？我现在是5080+64G 我在想要不要换128G内存 谢谢大佬</p>
]]></description><link>https://lcz.me/post/23639</link><guid isPermaLink="true">https://lcz.me/post/23639</guid><dc:creator><![CDATA[Marck Wong]]></dc:creator><pubDate>Fri, 09 Oct 2026 23:29:29 GMT</pubDate></item><item><title><![CDATA[Reply to i9-13900K + 128G + RTX5090 + Strata + MTP 部署  Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF on Fri, 09 Oct 2026 19:10:04 GMT]]></title><description><![CDATA[<p dir="auto">可以的，没有差距效果很棒<br />
<img src="https://upload.lcz.me/uploads/c7831ac1-442f-402d-9e92-8a1853be78ce.jpg" alt="2026_02188_.jpg" class="img-markdown  img-fluid" /><br />
<img src="https://upload.lcz.me/uploads/221b1a38-ee6d-4955-a284-171ccd30667f.jpg" alt="2026_02189_.jpg" class="img-markdown  img-fluid" /></p>
]]></description><link>https://lcz.me/post/23628</link><guid isPermaLink="true">https://lcz.me/post/23628</guid><dc:creator><![CDATA[明风影视mfys]]></dc:creator><pubDate>Fri, 09 Oct 2026 19:10:04 GMT</pubDate></item></channel></rss>