<?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[双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成]]></title><description><![CDATA[<h1>六张卡跑同一个 MiniMax-H3 视频工作流，差多少？</h1>
<blockquote>
<p dir="auto">机群里攒了几张不同世代的卡，一直想知道同一套 H3 Ref2VA 工作流跑下来到底差多少。<br />
花两天把 5090 / 4090 / 5080 / 170HX / 双 7900 XTX 全跑了一遍，记录冷热启动、显存、内存、功耗。<br />
数据都是实测，不是跑分软件估算。</p>
</blockquote>
<h2>测试条件</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>设置</th>
</tr>
</thead>
<tbody>
<tr>
<td>模型</td>
<td><code>minimax_h3_ref2va_pruned_int8_convrot</code></td>
</tr>
<tr>
<td>文本编码器</td>
<td><code>qwen3vl_32b_minimax_h3_nvfp4_awq</code></td>
</tr>
<tr>
<td>加速 LoRA</td>
<td><code>dmad_minimax_h3_8step_full_critic</code></td>
</tr>
<tr>
<td>步数</td>
<td><strong>8</strong>（与 LoRA 训练步数一致）</td>
</tr>
<tr>
<td>采样器</td>
<td>euler / simple</td>
</tr>
<tr>
<td>分辨率</td>
<td><strong>1344×768</strong>（16:9，1.032 MP）</td>
</tr>
<tr>
<td>ref_image_size</td>
<td><code>match</code></td>
</tr>
<tr>
<td>ComfyUI 显存策略</td>
<td>全部 <code>--lowvram</code></td>
</tr>
<tr>
<td>提示词</td>
<td>同一套三条（5s / 10s / 15s），带台词与音色参考</td>
</tr>
</tbody>
</table>
<h2>5 秒档</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>显卡</th>
<th>热启动</th>
<th>冷启动</th>
<th>显存峰值</th>
<th>功耗均值</th>
</tr>
</thead>
<tbody>
<tr>
<td>RTX 5090</td>
<td><strong>75 s</strong></td>
<td>75 s</td>
<td>30.9 GB</td>
<td>544 W</td>
</tr>
<tr>
<td>RTX 4090</td>
<td><strong>88 s</strong></td>
<td>98 s</td>
<td>42.4 GB</td>
<td>420 W</td>
</tr>
<tr>
<td>RTX 5080</td>
<td><strong>124 s</strong></td>
<td>134 s</td>
<td>15.2 GB</td>
<td>353 W</td>
</tr>
<tr>
<td>CMP 170HX</td>
<td><strong>219 s</strong></td>
<td>239 s</td>
<td>42.3 GB</td>
<td>226 W</td>
</tr>
<tr>
<td>7900 XTX ×2 (TP2)</td>
<td><strong>273 s</strong></td>
<td>625 s</td>
<td>32.5 GB</td>
<td>500 W</td>
</tr>
<tr>
<td>7900 XTX 单卡</td>
<td><strong>490 s</strong></td>
<td>717 s</td>
<td>17.7 GB</td>
<td>270 W</td>
</tr>
</tbody>
</table>
<h2>10 秒档</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>显卡</th>
<th>热启动</th>
<th>冷启动</th>
<th>显存峰值</th>
<th>功耗均值</th>
</tr>
</thead>
<tbody>
<tr>
<td>RTX 5090</td>
<td><strong>220 s</strong></td>
<td>224 s</td>
<td>30.9 GB</td>
<td>555 W</td>
</tr>
<tr>
<td>RTX 4090</td>
<td><strong>261 s</strong></td>
<td>264 s</td>
<td>46.5 GB</td>
<td>427 W</td>
</tr>
<tr>
<td>RTX 5080</td>
<td><strong>362 s</strong></td>
<td>367 s</td>
<td>15.2 GB</td>
<td>355 W</td>
</tr>
<tr>
<td>CMP 170HX</td>
<td><strong>710 s</strong></td>
<td>722 s</td>
<td>46.5 GB</td>
<td>235 W</td>
</tr>
<tr>
<td>7900 XTX ×2 (TP2)</td>
<td><strong>812 s</strong></td>
<td>1170 s</td>
<td>37.9 GB</td>
<td>502 W</td>
</tr>
<tr>
<td>7900 XTX 单卡</td>
<td><strong>1441 s</strong></td>
<td>1666 s</td>
<td>23.2 GB</td>
<td>271 W</td>
</tr>
</tbody>
</table>
<h2>15 秒档</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>显卡</th>
<th>热启动</th>
<th>冷启动</th>
<th>显存峰值</th>
<th>功耗均值</th>
</tr>
</thead>
<tbody>
<tr>
<td>RTX 5090</td>
<td><strong>404 s</strong></td>
<td>410 s</td>
<td>31.0 GB</td>
<td>561 W</td>
</tr>
<tr>
<td>RTX 4090</td>
<td><strong>477 s</strong></td>
<td>479 s</td>
<td>47.2 GB</td>
<td>434 W</td>
</tr>
<tr>
<td>RTX 5080</td>
<td><strong>665 s</strong></td>
<td>670 s</td>
<td>15.2 GB</td>
<td>356 W</td>
</tr>
<tr>
<td>CMP 170HX</td>
<td><strong>1367 s</strong></td>
<td>1371 s</td>
<td>50.2 GB</td>
<td>237 W</td>
</tr>
<tr>
<td>7900 XTX ×2 (TP2)</td>
<td><strong>1460 s</strong></td>
<td>1825 s</td>
<td>45.8 GB</td>
<td>502 W</td>
</tr>
<tr>
<td>7900 XTX 单卡</td>
<td><strong>OOM</strong></td>
<td>—</td>
<td>—</td>
<td>—</td>
</tr>
</tbody>
</table>
<h2>7900 XTX：双卡 TP2 对比单卡</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>视频时长</th>
<th>单卡热启动</th>
<th>双卡 TP2 热启动</th>
<th>对比</th>
</tr>
</thead>
<tbody>
<tr>
<td>5 秒</td>
<td>490 s</td>
<td>273 s</td>
<td>双卡快约 1.79 倍，耗时减少约 44%</td>
</tr>
<tr>
<td>10 秒</td>
<td>1441 s</td>
<td>812 s</td>
<td>双卡快约 1.77 倍，耗时减少约 44%</td>
</tr>
<tr>
<td>15 秒</td>
<td>OOM</td>
<td>1460 s</td>
<td>单卡显存不足，双卡可运行</td>
</tr>
</tbody>
</table>
<p dir="auto">5 秒档按显卡平均功耗估算，单卡约 36.8 Wh，双卡合计约 37.9 Wh：双卡节省了时间，单条显卡耗电量接近。以上为单次测量。</p>
<h2>能效：谁最省电</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>显卡</th>
<th>5 秒耗时</th>
<th>功耗均值</th>
<th>单条耗电</th>
</tr>
</thead>
<tbody>
<tr>
<td>RTX 4090</td>
<td>88 s</td>
<td>420 W</td>
<td><strong>10.3 Wh</strong></td>
</tr>
<tr>
<td>RTX 5090</td>
<td>75 s</td>
<td>544 W</td>
<td><strong>11.3 Wh</strong></td>
</tr>
<tr>
<td>RTX 5080</td>
<td>124 s</td>
<td>353 W</td>
<td><strong>12.2 Wh</strong></td>
</tr>
<tr>
<td>CMP 170HX</td>
<td>219 s</td>
<td>226 W</td>
<td><strong>13.7 Wh</strong></td>
</tr>
<tr>
<td>7900 XTX 单卡</td>
<td>490 s</td>
<td>270 W</td>
<td><strong>36.8 Wh</strong></td>
</tr>
<tr>
<td>7900 XTX ×2 (TP2)</td>
<td>273 s</td>
<td>500 W</td>
<td><strong>37.9 Wh</strong></td>
</tr>
</tbody>
</table>
<p dir="auto"><strong>4090 单条耗电最低（10.3 Wh），但速度不是最快。</strong></p>
<p dir="auto">按「每瓦出片量」算，5090 和 4090 接近（速度更快但功耗高），</p>
<p dir="auto">170HX 用 226 W 就跑完虽然慢，能耗只比 5090 高 21%，能耗表现不错。</p>
<p dir="auto"><strong>两张 7900 XTX 的能效最差</strong>——TP2 的通信开销和单卡的软件栈低效让它们耗电是 4090 的 3.5 倍。</p>
<h2>一个反直觉的发现：16 GB 能跑，24 GB 反而 OOM</h2>
<p dir="auto">RTX 5080 只有 16 GB，15 秒能跑完（665 s）。</p>
<p dir="auto">7900 XTX 有 24 GB，15 秒直接 <code>CUDA out of memory</code>。</p>
<p dir="auto">查下来原因不是容量，是<strong>能不能动态换页</strong>：</p>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>RTX 5080（16 GB 能跑）</th>
<th>7900 XTX（24 GB 不能跑）</th>
</tr>
</thead>
<tbody>
<tr>
<td><code>comfy-aimdo</code> 动态卸载</td>
<td><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/2705.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--white_check_mark" style="height:1.4em;width:auto;vertical-align:middle" title="✅" alt="✅" /> 有</td>
<td><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/274c.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--x" style="height:1.4em;width:auto;vertical-align:middle" title="❌" alt="❌" /> 无</td>
</tr>
<tr>
<td>20 GB 权重的处理</td>
<td>Staged 到系统内存，按需流式送入显存</td>
<td>必须全部常驻显存</td>
</tr>
<tr>
<td>显存峰值</td>
<td>15.21 GB（恒定满，持续换页）</td>
<td>23.98 GB（撑死）</td>
</tr>
<tr>
<td>系统内存参与</td>
<td>峰值 30.6 GB，起关键作用</td>
<td>不参与卸载</td>
</tr>
</tbody>
</table>
<p dir="auto">NVIDIA 的 <code>comfy-aimdo</code> 把显存变成了「缓存」而不是「仓库」，所以 16 GB 够用。</p>
<p dir="auto">AMD ROCm 缺这套机制，24 GB 是必须同时装下权重 + 激活的硬门槛。</p>
<p dir="auto"><strong>试过但无效的修法</strong>：<code>PYTORCH_HIP_ALLOC_CONF=expandable_segments:True</code></p>
<p dir="auto">碎片从 5.57 GB 降到 2.27 GB（确实生效），但总占用涨到 21.26 GB，还是差 1.19 GB。</p>
<p dir="auto">所以是硬性容量不足，不是碎片问题。</p>
<h2>结论</h2>
<ol>
<li><strong>要速度</strong> → RTX 5090（5 秒 75 s），三档全部领先</li>
<li><strong>要能效</strong> → RTX 4090（单条 10.3 Wh 最低）或 5090（11.3 Wh 但快 15%）</li>
<li><strong>要长视频</strong> → 5090 或 170HX，显存余量足</li>
<li><strong>不推荐</strong> 7900 XTX 双卡 TP2：15 秒 1460 s 比 170HX 单卡还慢</li>
<li><strong>降功耗策略</strong>：5080 可以压到 60%（只慢 4%）；4090 压到 85% 划算（省 14% 电只慢 4.5%），压到 60% 慢 42% 就不值了</li>
</ol>
<h2>测试环境</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>机器</th>
<th>配置</th>
</tr>
</thead>
<tbody>
<tr>
<td>5090</td>
<td>RTX 5090D 32G · Ryzen 7 9800X3D · 96 GB</td>
</tr>
<tr>
<td>170hx+4090</td>
<td>8× CMP 170HX + RTX 4090 48G · 2× EPYC 7542 · 512 GB</td>
</tr>
<tr>
<td>5080</td>
<td>RTX 5080 16G · Ryzen 5 9600X · 64 GB</td>
</tr>
<tr>
<td>7900xtx</td>
<td>2× RX 7900 XTX 24G · Ryzen 9 7900X · 192 GB</td>
</tr>
</tbody>
</table>
<p dir="auto"><strong>冷启动</strong> = 服务刚起的第一条（含模型加载 + 文本编码）</p>
<p dir="auto"><strong>热启动</strong> = 同服务紧接着的第二条（模型已驻显存，反映纯算力）</p>
]]></description><link>https://lcz.me/topic/2867</link><generator>RSS for Node</generator><lastBuildDate>Mon, 09 Nov 2026 05:21:07 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2867.rss" rel="self" type="application/rss+xml"/><pubDate>Thu, 05 Nov 2026 07:30:18 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 03:17:02 GMT]]></title><description><![CDATA[<p dir="auto"><img src="https://upload.lcz.me/uploads/34e2c17d-0171-45c9-9944-c387eee55079.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /><br />
求楼主给群众长长见识，这是个什么神仙坦克 怎么打的驱动 怎么控制的散热 ，从未见过如此神奇之物  锤哥也来说说@terry</p>
]]></description><link>https://lcz.me/post/32036</link><guid isPermaLink="true">https://lcz.me/post/32036</guid><dc:creator><![CDATA[范志刚]]></dc:creator><pubDate>Mon, 09 Nov 2026 03:17:02 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 03:12:23 GMT]]></title><description><![CDATA[<p dir="auto">“系统内存不参与” 这个结论不对，前段时间试过用1张7900xt跑H3模型int8权重，20G显存可以跑，系统内存肯定是参与卸载的</p>
]]></description><link>https://lcz.me/post/32033</link><guid isPermaLink="true">https://lcz.me/post/32033</guid><dc:creator><![CDATA[hippo]]></dc:creator><pubDate>Mon, 09 Nov 2026 03:12:23 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 03:05:03 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/terry" aria-label="Profile: terry">@<bdi>terry</bdi></a><br />
好的，我测试看一下。5080那台主机内存只有32G</p>
]]></description><link>https://lcz.me/post/32030</link><guid isPermaLink="true">https://lcz.me/post/32030</guid><dc:creator><![CDATA[懒人烘培]]></dc:creator><pubDate>Mon, 09 Nov 2026 03:05:03 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 03:03:54 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/data-wu" aria-label="Profile: Data-Wu">@<bdi>Data-Wu</bdi></a> <a href="/post/32023">说</a>:</p>
<p dir="auto">请问系统内存需要多大？5070ti 48G DDR5 内存能跑吗？谢谢</p>
</blockquote>
<p dir="auto">能跑就是慢一点，内存不够，就用SSD开虚拟内存。</p>
]]></description><link>https://lcz.me/post/32028</link><guid isPermaLink="true">https://lcz.me/post/32028</guid><dc:creator><![CDATA[perter]]></dc:creator><pubDate>Mon, 09 Nov 2026 03:03:54 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 02:57:58 GMT]]></title><description><![CDATA[<p dir="auto">请问系统内存需要多大？5070ti 48G DDR5 内存能跑吗？谢谢</p>
]]></description><link>https://lcz.me/post/32023</link><guid isPermaLink="true">https://lcz.me/post/32023</guid><dc:creator><![CDATA[Data Wu]]></dc:creator><pubDate>Mon, 09 Nov 2026 02:57:58 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Mon, 09 Nov 2026 02:57:02 GMT]]></title><description><![CDATA[<p dir="auto">CK+sol-attn 得用上啊</p>
]]></description><link>https://lcz.me/post/32022</link><guid isPermaLink="true">https://lcz.me/post/32022</guid><dc:creator><![CDATA[伊明]]></dc:creator><pubDate>Mon, 09 Nov 2026 02:57:02 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 12:48:16 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/xiongjie-gao" aria-label="Profile: xiongjie-gao">@<bdi>xiongjie-gao</bdi></a> 还得我弟支撑，我忙完了才有空测试。</p>
]]></description><link>https://lcz.me/post/31948</link><guid isPermaLink="true">https://lcz.me/post/31948</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Sun, 08 Nov 2026 12:48:16 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 12:27:09 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/terry" aria-label="Profile: terry">@<bdi>terry</bdi></a> 好的哇，跑完这个视频安排</p>
]]></description><link>https://lcz.me/post/31940</link><guid isPermaLink="true">https://lcz.me/post/31940</guid><dc:creator><![CDATA[xiongjie gao]]></dc:creator><pubDate>Sun, 08 Nov 2026 12:27:09 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 12:21:18 GMT]]></title><description><![CDATA[<p dir="auto"><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/1f525.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--fire" style="height:1.4em;width:auto;vertical-align:middle" title="🔥" alt="🔥" /> 本帖参与讨论人数已达 <strong>10+</strong>，入选<strong>热议</strong>，作者 <strong>+3 积分</strong>奖励！</p>
<blockquote>
<p dir="auto">奖励凭证：热议 +3 分 · 编号 h2867-3611-1</p>
</blockquote>
]]></description><link>https://lcz.me/post/31938</link><guid isPermaLink="true">https://lcz.me/post/31938</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Sun, 08 Nov 2026 12:21:18 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 12:26:45 GMT]]></title><description><![CDATA[<p dir="auto">非常不错的测试，可以剽一期视频。</p>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>RTX 5080（16 GB 能跑）</th>
<th>7900 XTX（24 GB 不能跑）</th>
</tr>
</thead>
<tbody>
<tr>
<td><code>comfy-aimdo</code> 动态卸载</td>
<td><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/2705.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--white_check_mark" style="height:1.4em;width:auto;vertical-align:middle" title="✅" alt="✅" /> 有</td>
<td><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/274c.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--x" style="height:1.4em;width:auto;vertical-align:middle" title="❌" alt="❌" /> 无</td>
</tr>
<tr>
<td>20 GB 权重的处理</td>
<td>Staged 到系统内存，按需流式送入显存</td>
<td>必须全部常驻显存</td>
</tr>
<tr>
<td>显存峰值</td>
<td>15.21 GB（恒定满，持续换页）</td>
<td>23.98 GB（撑死）</td>
</tr>
<tr>
<td>系统内存参与</td>
<td>峰值 30.6 GB，起关键作用</td>
<td>不参与卸载</td>
</tr>
</tbody>
</table>
<p dir="auto">GPT表示表示不属实，是没配置好<br />
<a class="plugin-mentions-user plugin-mentions-a" href="/user/xiongjie-gao" aria-label="Profile: xiongjie-gao">@<bdi>xiongjie-gao</bdi></a><br />
<a class="plugin-mentions-user plugin-mentions-a" href="/user/%E6%87%92%E4%BA%BA%E7%83%98%E5%9F%B9" aria-label="Profile: 懒人烘培">@<bdi>懒人烘培</bdi></a><br />
我弟可以验证下。</p>
]]></description><link>https://lcz.me/post/31937</link><guid isPermaLink="true">https://lcz.me/post/31937</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Sun, 08 Nov 2026 12:26:45 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 12:20:02 GMT]]></title><description><![CDATA[<p dir="auto"><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/2b50.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--star" style="height:1.4em;width:auto;vertical-align:middle" title="⭐" alt="⭐" /> 本帖由<strong>优秀</strong>升级为<strong>精华</strong>（两者互斥，只补差价），作者 <strong>+3 积分</strong>，累计 6 分。</p>
<blockquote>
<p dir="auto">补差凭证：优秀升精华 +3 分 · 编号 ef2867-3611-1</p>
</blockquote>
]]></description><link>https://lcz.me/post/31936</link><guid isPermaLink="true">https://lcz.me/post/31936</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Sun, 08 Nov 2026 12:20:02 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sun, 08 Nov 2026 01:55:51 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/greybid" aria-label="Profile: greybid">@<bdi>greybid</bdi></a> 确实是，如果只是为了MINIMAX H3那可以不用考虑7900xtx了，我准备双卡留着跑qwen4 flash</p>
]]></description><link>https://lcz.me/post/31838</link><guid isPermaLink="true">https://lcz.me/post/31838</guid><dc:creator><![CDATA[Big Liao]]></dc:creator><pubDate>Sun, 08 Nov 2026 01:55:51 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Sat, 07 Nov 2026 16:52:44 GMT]]></title><description><![CDATA[<p dir="auto">今天折腾了一下，单卡7900XTX，跑4步的1344x768的10秒时长视频耗时764秒，15秒时长视频耗时只要1229秒，就是感觉画质有点点烂。另外感觉高强度跑视频生成的工况下，机器放在身边的同学，建议不要买涡轮卡，那是真的吵啊。</p>
]]></description><link>https://lcz.me/post/31814</link><guid isPermaLink="true">https://lcz.me/post/31814</guid><dc:creator><![CDATA[Michael ZHANG]]></dc:creator><pubDate>Sat, 07 Nov 2026 16:52:44 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Fri, 06 Nov 2026 10:47:04 GMT]]></title><description><![CDATA[<p dir="auto">隔壁帖子不是说双7900xtx TP连接效率是4090的75%吗？为何实测跑minimax h3的效率还不到5080的30%？<br />
这么看简直拉完了啊</p>
]]></description><link>https://lcz.me/post/31586</link><guid isPermaLink="true">https://lcz.me/post/31586</guid><dc:creator><![CDATA[greybid]]></dc:creator><pubDate>Fri, 06 Nov 2026 10:47:04 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Fri, 06 Nov 2026 10:01:36 GMT]]></title><description><![CDATA[<p dir="auto">minimax h3 工作流= 文本理解（脱裤子）+生成视频（拉屎），这位大佬扯得是完整工作流，之前烘培大佬测的是生成视频部分。通俗的讲就是 n卡脱裤子+拉屎速度都快，a卡脱裤子特别慢，但拉屎特别快。我寻思，如果让一个5070ti专门负责脱裤子，双7900xtx负责拉屎。那效率不就上来了</p>
]]></description><link>https://lcz.me/post/31580</link><guid isPermaLink="true">https://lcz.me/post/31580</guid><dc:creator><![CDATA[yuyuan chen]]></dc:creator><pubDate>Fri, 06 Nov 2026 10:01:36 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 20:11:19 GMT]]></title><description><![CDATA[<p dir="auto"><img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/2705.png?v=5be5614064c" class="not-responsive emoji emoji-android emoji--white_check_mark" style="height:1.4em;width:auto;vertical-align:middle" title="✅" alt="✅" /> 版主审定：本帖设为<strong>优秀</strong>，作者 <strong>+3 积分</strong>奖励，继续加油！</p>
<blockquote>
<p dir="auto">奖励凭证：优秀 +3 分 · 编号 e2867-3611-1</p>
</blockquote>
]]></description><link>https://lcz.me/post/31421</link><guid isPermaLink="true">https://lcz.me/post/31421</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Thu, 05 Nov 2026 20:11:19 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 11:08:45 GMT]]></title><description><![CDATA[<p dir="auto">有没有测试过 R9700</p>
]]></description><link>https://lcz.me/post/31341</link><guid isPermaLink="true">https://lcz.me/post/31341</guid><dc:creator><![CDATA[大圣]]></dc:creator><pubDate>Thu, 05 Nov 2026 11:08:45 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 10:59:02 GMT]]></title><description><![CDATA[<p dir="auto">7900必须要int8直出，amd速度并不慢，但是需要设置好，加速插件也要搭配好</p>
]]></description><link>https://lcz.me/post/31339</link><guid isPermaLink="true">https://lcz.me/post/31339</guid><dc:creator><![CDATA[fcys]]></dc:creator><pubDate>Thu, 05 Nov 2026 10:59:02 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 10:29:01 GMT]]></title><description><![CDATA[<p dir="auto">没想到就算双卡速度上差距跟英伟达比还是很大。</p>
]]></description><link>https://lcz.me/post/31329</link><guid isPermaLink="true">https://lcz.me/post/31329</guid><dc:creator><![CDATA[Kai Liu]]></dc:creator><pubDate>Thu, 05 Nov 2026 10:29:01 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 09:01:23 GMT]]></title><description><![CDATA[<p dir="auto">老黄的卡买不了吃亏买不了上当,贵有贵的道理.CUDA深耕多年护城河依旧很难跨越.</p>
]]></description><link>https://lcz.me/post/31318</link><guid isPermaLink="true">https://lcz.me/post/31318</guid><dc:creator><![CDATA[Goitex]]></dc:creator><pubDate>Thu, 05 Nov 2026 09:01:23 GMT</pubDate></item><item><title><![CDATA[Reply to 双卡7900xtx vs 170hx vs 5090d32gb vs 4090 48gb vs 5080视频生成 on Thu, 05 Nov 2026 07:35:56 GMT]]></title><description><![CDATA[<p dir="auto">DMAD 8步 加速lora是我测试里最近比较流行的加速方式里效果最好的，pika也不错，不过声音和画面不太喜欢，但是动作遵从度最强。<br />
1mp分辨率是minimax训练的默认分辨率，没有必要直接生成1080p甚至更高，如果需要更高画质，用DLSS5 增强一下就挺好的了，如果要求更高用latent upscale或者新出的Open-H3-Regenerate-2K也可以。DLSS5最快，都生成好了，统一发给5090 ，5秒1mp视频大约12秒就能upscale到2k。</p>
]]></description><link>https://lcz.me/post/31296</link><guid isPermaLink="true">https://lcz.me/post/31296</guid><dc:creator><![CDATA[Big Liao]]></dc:creator><pubDate>Thu, 05 Nov 2026 07:35:56 GMT</pubDate></item></channel></rss>