<?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[山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。]]></title><description><![CDATA[<p dir="auto"><img src="https://upload.lcz.me/uploads/d4f10b83-a4d4-49a0-97c2-2c1c039192d4.png" alt="ScreenShot_2026-05-27_142430_657.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/21cf44f0-0971-40db-a4ff-276a48c27086.png" alt="ScreenShot_2026-05-27_212415_552.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/b9f819f5-b433-4d50-8e63-240363564083.png" alt="ScreenShot_2026-05-27_142655_840.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/024cfedb-5439-4cd7-b8af-90e99b02c2bb.png" alt="ScreenShot_2026-06-01_225729_777.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/5481d7bb-8688-4543-af88-9952cdd6897b.png" alt="ScreenShot_2026-06-01_225840_231.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/247d8f5f-5539-49d8-9104-85750086065a.png" alt="ScreenShot_2026-06-01_230507_177.png" class=" img-fluid img-markdown" /> <img src="https://upload.lcz.me/uploads/deaca217-3280-4e73-b57a-a851a54169da.png" alt="ScreenShot_2026-06-01_230535_539.png" class=" img-fluid img-markdown" /></p>
<p dir="auto">#!/bin/bash<br />
export LD_LIBRARY_PATH=/home/qwe/llama.cpp/build/bin:$LD_LIBRARY_PATH</p>
<p dir="auto">MODEL=/home/qwe/models/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP.gguf<br />
PORT=8081<br />
LLAMA_SERVER=/home/qwe/llama.cpp/build/bin/llama-server</p>
<p dir="auto">exec "$LLAMA_SERVER" <br />
--n-predict 16384 <br />
--fit off <br />
--split-mode tensor --tensor-split 1,1 <br />
--device CUDA0,CUDA1 <br />
-m "$MODEL" <br />
--host 0.0.0.0 --port "$PORT" <br />
-t 0 -ngl 99 -np 1 <br />
--no-mmap <br />
--kv-unified --flash-attn on --ctx-size 160000 <br />
--spec-type draft-mtp --spec-draft-n-max 2 <br />
--repeat-penalty 1.1 <br />
--min-p 0.02 <br />
--temp 0.6 --top-k 20 --top-p 0.95</p>
]]></description><link>https://lcz.me/topic/382/山寨x99主板-32g-ddr3内存-两张5060ti-16g-llama.cpp-qwen3.6-27b-nvfp4版-40-70t-s-现在够用未来会更好</link><generator>RSS for Node</generator><lastBuildDate>Wed, 01 Jul 2026 12:08:36 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/382.rss" rel="self" type="application/rss+xml"/><pubDate>Mon, 01 Jun 2026 15:06:31 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sun, 07 Jun 2026 01:01:25 GMT]]></title><description><![CDATA[<p dir="auto">测试数据非常有参考意义，置顶，有prefill速度可以发下，但影响不是很大。</p>
]]></description><link>https://lcz.me/post/5414</link><guid isPermaLink="true">https://lcz.me/post/5414</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Sun, 07 Jun 2026 01:01:25 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sun, 07 Jun 2026 03:18:44 GMT]]></title><description><![CDATA[<p dir="auto">Prefill 速度？</p>
<p dir="auto">刚刚问ai prefill 大概有1000t/s 也不错<br />
主要是价钱便宜</p>
]]></description><link>https://lcz.me/post/5411</link><guid isPermaLink="true">https://lcz.me/post/5411</guid><dc:creator><![CDATA[applejuice]]></dc:creator><pubDate>Sun, 07 Jun 2026 03:18:44 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sat, 06 Jun 2026 23:41:11 GMT]]></title><description><![CDATA[<p dir="auto">华南x99 f8hplus主板，双显卡一插上就启动不了，Above 4G Decoding：Enabled也已打开。大神有没有好的方法，已经试错了两天了，想吐了。</p>
]]></description><link>https://lcz.me/post/5407</link><guid isPermaLink="true">https://lcz.me/post/5407</guid><dc:creator><![CDATA[yzl8850622]]></dc:creator><pubDate>Sat, 06 Jun 2026 23:41:11 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sat, 06 Jun 2026 22:13:58 GMT]]></title><description><![CDATA[<p dir="auto">@Gang Cheng 我来补充一下两个方案的具体对比，帮你做决定：</p>
<p dir="auto"><strong>方案A：再加一张5060Ti 16G（双卡）</strong><br />
优势：</p>
<ul>
<li>总显存32GB，比7900XTX多8GB，跑大模型更从容</li>
<li>Blackwell架构的NVFP4是杀手锏——Qwen3.6 27B用NVFP4量化只需要约17GB，一张卡就能跑，双卡甚至可以跑更大的模型</li>
<li>llama.cpp双卡tensor parallelism效率很高，实测双5060Ti跑27B能有40-70T/s</li>
<li>成本低：再买一张5060Ti约2500-3000元</li>
</ul>
<p dir="auto">劣势：</p>
<ul>
<li>双卡需要主板有两条PCIe x16槽，电源要够</li>
<li>ComfyUI等生图场景分卡有额外开销</li>
</ul>
<p dir="auto"><strong>方案B：换7900XTX 24G</strong><br />
优势：</p>
<ul>
<li>单卡24GB，ROCm生态对vLLM/SGLang支持好</li>
<li>生图/视频场景（ComfyUI）单卡不需要分卡，更省心</li>
<li>单卡推理吞吐比单张5060Ti高</li>
<li>保修还有2年</li>
</ul>
<p dir="auto">劣势：</p>
<ul>
<li>总显存反而比双5060Ti少8GB</li>
<li>不支持NVFP4量化</li>
<li>卖卡+买卡差价大，综合成本更高</li>
</ul>
<p dir="auto"><strong>我的建议：</strong> 如果你主要跑llama.cpp纯推理，加一张5060Ti双卡是更优解——32GB总显存+NVFP4，性价比很高。如果你未来主要跑ComfyUI生图/视频，那7900XTX的24G单卡更省心。两个都要兼顾的话，建议先加5060Ti双卡试试，32GB显存是想换也换不来的硬优势。</p>
]]></description><link>https://lcz.me/post/5406</link><guid isPermaLink="true">https://lcz.me/post/5406</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Sat, 06 Jun 2026 22:13:58 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sat, 06 Jun 2026 15:22:02 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/gang-cheng" aria-label="Profile: Gang-Cheng">@<bdi>Gang-Cheng</bdi></a>  7900xtx  / R9700  VRAM 帶來的改善比較大</p>
]]></description><link>https://lcz.me/post/5389</link><guid isPermaLink="true">https://lcz.me/post/5389</guid><dc:creator><![CDATA[CS6]]></dc:creator><pubDate>Sat, 06 Jun 2026 15:22:02 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Sat, 06 Jun 2026 15:02:37 GMT]]></title><description><![CDATA[<p dir="auto">我现在有一张5060ti16g我是想换一张7900xtx好还是再加一张5060ti16g呢？麻烦大神给我指导一下~</p>
]]></description><link>https://lcz.me/post/5384</link><guid isPermaLink="true">https://lcz.me/post/5384</guid><dc:creator><![CDATA[Gang Cheng]]></dc:creator><pubDate>Sat, 06 Jun 2026 15:02:37 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Wed, 03 Jun 2026 15:46:29 GMT]]></title><description><![CDATA[<p dir="auto">看到5060 Ti 16GB 性價比就是高 ～ 總是要讚一個 ：）</p>
]]></description><link>https://lcz.me/post/4865</link><guid isPermaLink="true">https://lcz.me/post/4865</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Wed, 03 Jun 2026 15:46:29 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Mon, 01 Jun 2026 19:39:30 GMT]]></title><description><![CDATA[<p dir="auto">非常好的分享，关键参数用文字贴下。</p>
]]></description><link>https://lcz.me/post/4595</link><guid isPermaLink="true">https://lcz.me/post/4595</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Mon, 01 Jun 2026 19:39:30 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Mon, 01 Jun 2026 15:44:12 GMT]]></title><description><![CDATA[<p dir="auto">50系列能用NVFP是厉害啊！表现比双3080-20G还好</p>
]]></description><link>https://lcz.me/post/4578</link><guid isPermaLink="true">https://lcz.me/post/4578</guid><dc:creator><![CDATA[comeN]]></dc:creator><pubDate>Mon, 01 Jun 2026 15:44:12 GMT</pubDate></item><item><title><![CDATA[Reply to 山寨X99主板，32G DDR3内存，两张5060TI 16G llama.cpp Qwen3.6 27B NVFP4版 40-70T&#x2F;S 现在够用未来会更好。 on Mon, 01 Jun 2026 15:12:44 GMT]]></title><description><![CDATA[<p dir="auto">模型的部署用跑在deepseek下的Hermes agent 自动安装部署的，直接把模型网页扔给他，让他学习参考抄作业。</p>
]]></description><link>https://lcz.me/post/4576</link><guid isPermaLink="true">https://lcz.me/post/4576</guid><dc:creator><![CDATA[asd2667]]></dc:creator><pubDate>Mon, 01 Jun 2026 15:12:44 GMT</pubDate></item></channel></rss>