<?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[[7900XTX] Minimax H3视频生成速度测试报告]]></title><description><![CDATA[<h2>1. 实测参数与时间</h2>
<p dir="auto">我的工作流没有用任何的加速节点，直接选择10步采用就感觉很快了。用其他加速节点估计是没有匹配Linux或者ROCm的版本，采样后都是花屏。我的机器直出10s太勉强，用5s还是舒适一些。或者是低分辨率换其他方式超分放大。有兴趣的人可以进一步研究。</p>
<p dir="auto">对我来说目前是够用了。我主要用来做游戏的512x512 , 2s的角色动画，用来抽帧， 平均90s一个小视频，批量抽卡也很快。</p>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>时长</th>
<th>分辨率</th>
<th>采样步数</th>
<th>耗时</th>
</tr>
</thead>
<tbody>
<tr>
<td>5s</td>
<td>288x576</td>
<td>10</td>
<td>138.93s</td>
</tr>
<tr>
<td>10s</td>
<td>288x576</td>
<td>10</td>
<td>196.60s</td>
</tr>
<tr>
<td>5s</td>
<td>1024x588</td>
<td>10</td>
<td>335.71s</td>
</tr>
<tr>
<td>10s</td>
<td>1024x588</td>
<td>10</td>
<td>778.87s</td>
</tr>
</tbody>
</table>
<h2>** 我把我本机的配置环境，全量显示出来，如果有看不懂的，拿去喂AI就好了 =.= **</h2>
<h2>2. 概览</h2>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>内容</th>
</tr>
</thead>
<tbody>
<tr>
<td>主机名</td>
<td>ferychen7900XTX</td>
</tr>
<tr>
<td>操作系统</td>
<td>CachyOS（Arch Linux 系）</td>
</tr>
<tr>
<td>内核</td>
<td>Linux 7.1.8-1-cachyos（x86_64, PREEMPT_DYNAMIC）</td>
</tr>
<tr>
<td>主板</td>
<td>ASUS TUF GAMING B760M-PLUS D4</td>
</tr>
<tr>
<td>CPU</td>
<td>Intel Core i5-13490F（13 代 Raptor Lake）</td>
</tr>
<tr>
<td>内存</td>
<td>38 GiB（另有 51 GiB 交换分区）</td>
</tr>
<tr>
<td>GPU</td>
<td>AMD Radeon RX 7900 XTX（Navi 31 / gfx1100，24 GiB 显存）</td>
</tr>
<tr>
<td>计算栈</td>
<td>ROCm 7.14 + PyTorch 2.11.0+rocm7.14.0</td>
</tr>
<tr>
<td>Python 环境</td>
<td>Python 3.12.13</td>
</tr>
</tbody>
</table>
<hr />
<h2>3. 操作系统与内核</h2>
<ul>
<li>发行版：CachyOS（ID=cachyos, ID_LIKE=arch），滚动更新</li>
<li>内核：7.1.8-1-cachyos（CachyOS 定制内核，含性能调度优化）</li>
<li>桌面 / 显示服务器：Hyprland（Wayland，UWSM 启动）</li>
<li>登录管理器：greetd + noctalia-greeter</li>
</ul>
<hr />
<h2>4. 硬件配置</h2>
<h3>4.1 CPU</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>规格</th>
</tr>
</thead>
<tbody>
<tr>
<td>型号</td>
<td>13th Gen Intel Core i5-13490F</td>
</tr>
<tr>
<td>核心 / 线程</td>
<td>10 核 16 线程（6P + 4E）</td>
</tr>
<tr>
<td>频率</td>
<td>基准 ~2.5 GHz，最大睿频 4.8 GHz</td>
</tr>
<tr>
<td>缓存</td>
<td>L3 24 MiB，L2 9.5 MiB</td>
</tr>
<tr>
<td>虚拟化</td>
<td>VT-x 支持</td>
</tr>
</tbody>
</table>
<h3>4.2 内存</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>规格</th>
</tr>
</thead>
<tbody>
<tr>
<td>物理内存</td>
<td>38 GiB（DDR4）</td>
</tr>
<tr>
<td>交换</td>
<td>51 GiB</td>
</tr>
</tbody>
</table>
<h3>4.3 GPU（AI 推理主力）</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>项目</th>
<th>规格</th>
</tr>
</thead>
<tbody>
<tr>
<td>型号</td>
<td>AMD Radeon RX 7900 XTX</td>
</tr>
<tr>
<td>核心</td>
<td>Navi 31（gfx1100, RDNA 3）</td>
</tr>
<tr>
<td>显存</td>
<td>24 GiB GDDR6（实际 25.7 GB，rocm-smi 读数）</td>
</tr>
<tr>
<td>功耗上限</td>
<td>303 W</td>
</tr>
</tbody>
</table>
<hr />
<h2>5. GPU 计算栈（ROCm）</h2>
<ul>
<li>ROCm 版本：7.14（随 PyTorch wheel 绑定）；rocm-smi-lib 7.8.0</li>
<li>内核模块：amdgpu、amdkfd 已加载</li>
<li>检测命令：<code>rocm-smi</code>（可查看温度、功耗、显存占用）</li>
<li>常见坑位备注：pt-2.12(rocm7.14) 花屏 → 清 <code>~/.triton/cache</code> + <code>~/.cache/comgr</code> 即恢复</li>
</ul>
<hr />
<h2>6. Python 虚拟环境（pt-2.11-r714）</h2>
<ul>
<li>Python：3.12.13</li>
<li>用途：ComfyUI 宿主环境，本地运行 MiniMax H3 视频生成工作流（含 Ref2VA 全参考视频模式）</li>
<li>包总量：约 188 个</li>
</ul>
<h3>6.1 核心包版本</h3>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>包</th>
<th>版本</th>
</tr>
</thead>
<tbody>
<tr>
<td>torch</td>
<td>2.11.0+rocm7.14.0</td>
</tr>
<tr>
<td>torchvision</td>
<td>0.26.0+rocm7.14.0</td>
</tr>
<tr>
<td>torchaudio</td>
<td>2.11.0+rocm7.14.0</td>
</tr>
<tr>
<td>transformers</td>
<td>5.14.1</td>
</tr>
<tr>
<td>diffusers</td>
<td>0.27.2</td>
</tr>
<tr>
<td>safetensors</td>
<td>0.8.0</td>
</tr>
<tr>
<td>numpy</td>
<td>2.4.4</td>
</tr>
<tr>
<td>pillow</td>
<td>12.2.0</td>
</tr>
<tr>
<td>sentencepiece</td>
<td>0.2.2</td>
</tr>
<tr>
<td>modelscope</td>
<td>1.39.0</td>
</tr>
<tr>
<td>huggingface_hub</td>
<td>1.27.0</td>
</tr>
<tr>
<td>torchsde</td>
<td>0.2.6</td>
</tr>
</tbody>
</table>
<h3>6.2 ComfyUI 相关</h3>
<ul>
<li>comfyui_frontend_package 1.48.7</li>
<li>comfyui-manager 4.2.2</li>
<li>comfyui_workflow_templates 0.11.41（及配套 media 模板包）</li>
</ul>
<h3>6.3 PyTorch 运行时验证结果</h3>
<pre><code>torch:      2.11.0+rocm7.14.0
hip:        7.14.60850
cuda avail: True（走 HIP 后端）
device:     AMD Radeon RX 7900 XTX
</code></pre>
<hr />
<p dir="auto"><img src="https://upload.lcz.me/uploads/9ae20079-2f7c-4ce3-a7fc-6edd3763345d.jpeg" alt="4e573c77-e6fe-421f-b044-27df4eaa0c5b-image.jpeg" class=" img-fluid img-markdown" /></p>
]]></description><link>https://lcz.me/topic/1160/7900xtx-minimax-h3视频生成速度测试报告</link><generator>RSS for Node</generator><lastBuildDate>Sat, 22 Aug 2026 03:26:48 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/1160.rss" rel="self" type="application/rss+xml"/><pubDate>Mon, 17 Aug 2026 09:59:40 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to [7900XTX] Minimax H3视频生成速度测试报告 on Mon, 17 Aug 2026 12:25:58 GMT]]></title><description><![CDATA[<p dir="auto">非常好老弟，这个帖子看起来就赏心悦目，效果其实不用太追求，现在的视频竞争还是内容大于形式，重要是点子。</p>
]]></description><link>https://lcz.me/post/12525</link><guid isPermaLink="true">https://lcz.me/post/12525</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Mon, 17 Aug 2026 12:25:58 GMT</pubDate></item></channel></rss>