<?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[Basalt專武N卡50系列引擎 Qwen3.8-Flash-Next on NVIDIA Blackwell]]></title><description><![CDATA[<p dir="auto">Strata 簡直是引領風潮<br />
後面一堆新花樣都出來了<br />
Basalt 這支援 NVIDIA Blackwell 架構<br />
<a href="https://github.com/jesdga95/basalt" rel="nofollow ugc">https://github.com/jesdga95/basalt</a><br />
上面的測試報告如下<br />
<img src="https://upload.lcz.me/uploads/4a21b867-3bf6-417e-8c32-17483b67ddca.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<p dir="auto">然後我自己照做的效率如下(QFN nvfp4 , 5090x2 , 64G RAM)<br />
<img src="https://upload.lcz.me/uploads/f7b87f6a-1ed1-4047-8302-806dcd19a927.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /><br />
<img src="https://upload.lcz.me/uploads/1efb9432-43ca-40ca-b580-cdd3e428b2d0.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<p dir="auto">大致上是對的上的</p>
<p dir="auto">目前這些模型的短板都是多功不大行（不像是稠密模型多併發會有總輸出增加的紅利）</p>
<p dir="auto">適合單輸出玩家</p>
]]></description><link>https://lcz.me/topic/2763</link><generator>RSS for Node</generator><lastBuildDate>Mon, 09 Nov 2026 09:24:40 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2763.rss" rel="self" type="application/rss+xml"/><pubDate>Sun, 01 Nov 2026 06:32:24 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to Basalt專武N卡50系列引擎 Qwen3.8-Flash-Next on NVIDIA Blackwell on Mon, 02 Nov 2026 09:11:28 GMT]]></title><description><![CDATA[<p dir="auto">吞吐太可怕了....<br />
但其實高併發噴不上去 這不是模型的問題 我用gufo跑qfn一樣四路可以噴到2.5倍<br />
是strata推論框架的問題，沒辦法 他要搞好異構運算與冷熱交換 高併發是被犧牲的<br />
所以我一直認為不要砸太多錢在strata上~ 夠跑就好</p>
]]></description><link>https://lcz.me/post/30521</link><guid isPermaLink="true">https://lcz.me/post/30521</guid><dc:creator><![CDATA[dardeaw feng]]></dc:creator><pubDate>Mon, 02 Nov 2026 09:11:28 GMT</pubDate></item><item><title><![CDATA[Reply to Basalt專武N卡50系列引擎 Qwen3.8-Flash-Next on NVIDIA Blackwell on Mon, 02 Nov 2026 08:30:30 GMT]]></title><description><![CDATA[<p dir="auto">确实厉害，5090直接起飞</p>
]]></description><link>https://lcz.me/post/30514</link><guid isPermaLink="true">https://lcz.me/post/30514</guid><dc:creator><![CDATA[大圣]]></dc:creator><pubDate>Mon, 02 Nov 2026 08:30:30 GMT</pubDate></item><item><title><![CDATA[Reply to Basalt專武N卡50系列引擎 Qwen3.8-Flash-Next on NVIDIA Blackwell on Mon, 02 Nov 2026 08:04:16 GMT]]></title><description><![CDATA[<p dir="auto">谁可以试一下使用双卡5060ti或5070ti跑跑这框架</p>
]]></description><link>https://lcz.me/post/30512</link><guid isPermaLink="true">https://lcz.me/post/30512</guid><dc:creator><![CDATA[李野火]]></dc:creator><pubDate>Mon, 02 Nov 2026 08:04:16 GMT</pubDate></item><item><title><![CDATA[Reply to Basalt專武N卡50系列引擎 Qwen3.8-Flash-Next on NVIDIA Blackwell on Mon, 02 Nov 2026 02:18:45 GMT]]></title><description><![CDATA[<p dir="auto">感謝大大分享<br />
好多推理引擎 我都快記不得誰跟誰了<br />
目前只記得 llama.cpp, SGLang, vLLM, Strata 現在多了一個Basalt (玄武岩)</p>
<p dir="auto">看來多併發方面, 稠密模型 27B 還是有它的優勢在</p>
<p dir="auto">這Basalt 用來作單發專案Coding, 看來效率會很驚人</p>
]]></description><link>https://lcz.me/post/30432</link><guid isPermaLink="true">https://lcz.me/post/30432</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Mon, 02 Nov 2026 02:18:45 GMT</pubDate></item></channel></rss>