<?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[接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash]]></title><description><![CDATA[<p dir="auto">接了一个装本地AI的活，苹果Studio 512G统一内存，M3 Max ，跑Deepseek V4 flash<br />
可能需要折腾一下<br />
如果顺利，<br />
会把截图和过程放出来。<br />
有人知道ds4.c 这个架构吗？</p>
]]></description><link>https://lcz.me/topic/124/接了一个装本地ai的活-苹果studio-512g统一内存-跑deepseek-v4-flash</link><generator>RSS for Node</generator><lastBuildDate>Wed, 20 May 2026 07:04:20 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/124.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 13 May 2026 08:58:51 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 13:46:40 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/devin-hi" aria-label="Profile: Devin-Hi">@<bdi>Devin-Hi</bdi></a> 单独发给帖子，多弄几张图谈谈真实感受，给我做一期视频，云下，这玩意我可能买不起了....</p>
]]></description><link>https://lcz.me/post/1636</link><guid isPermaLink="true">https://lcz.me/post/1636</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Thu, 14 May 2026 13:46:40 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 13:45:24 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/grayson-ren" aria-label="Profile: Grayson-Ren">@<bdi>Grayson-Ren</bdi></a></p>
<p dir="auto">从我的角度来说，你跑一个大模型还是几个大模型，你会发现GPU就是100%了，但内存就是30%。 就是这样，等待的时间都是GPU的处理时间。</p>
]]></description><link>https://lcz.me/post/1635</link><guid isPermaLink="true">https://lcz.me/post/1635</guid><dc:creator><![CDATA[Devin Hi]]></dc:creator><pubDate>Thu, 14 May 2026 13:45:24 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 13:41:27 GMT]]></title><description><![CDATA[<p dir="auto">稍后发图<br />
是的，我觉得赶紧卖，一台机子不到4个月，赚了大几万，不到十万买的，说最高能卖20万，有点炒币的感觉了</p>
]]></description><link>https://lcz.me/post/1629</link><guid isPermaLink="true">https://lcz.me/post/1629</guid><dc:creator><![CDATA[Devin Hi]]></dc:creator><pubDate>Thu, 14 May 2026 13:41:27 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 13:37:19 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/devin-hi" aria-label="Profile: Devin-Hi">@<bdi>Devin-Hi</bdi></a> 现在内存缺货，显得M3 Urtral很值钱，事实上它真的不行，早点卖个好价钱，换RTX Pro 6000.</p>
]]></description><link>https://lcz.me/post/1622</link><guid isPermaLink="true">https://lcz.me/post/1622</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Thu, 14 May 2026 13:37:19 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 13:35:20 GMT]]></title><description><![CDATA[<p dir="auto">装机完毕，先说结论：M3 utral  512G，内存的确豪横，可以同时跑：deepseek V4 flash  （q2量化） 和 Qwen 3.6 -27B 稠密模型，体验30 t/秒， 同时还跑了小龙虾和 hermes，内存占用率30%左右，GPU拉满，CPU 40%左右。第一次看到 一台设备 是内存处于闲置状态。感觉 M3 256G内存足够了，再高就是闲置，目前一台价格等于一台车。。。。。。。穷人看着眼馋，说卖了能换好几个Pro 6000 和 4090 呢。效果不如云端deepseek V4 flash。对于在乎成本的人来说真的没有必要。当然王思聪一类的土老板，可以玩具，发热不高，比我的7900XTX 冷静多了。</p>
]]></description><link>https://lcz.me/post/1620</link><guid isPermaLink="true">https://lcz.me/post/1620</guid><dc:creator><![CDATA[Devin Hi]]></dc:creator><pubDate>Thu, 14 May 2026 13:35:20 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 10:26:55 GMT]]></title><description><![CDATA[<p dir="auto">我在部署ds4引擎 等测试下结果看速度如何</p>
]]></description><link>https://lcz.me/post/1563</link><guid isPermaLink="true">https://lcz.me/post/1563</guid><dc:creator><![CDATA[Grayson Ren]]></dc:creator><pubDate>Thu, 14 May 2026 10:26:55 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 06:04:59 GMT]]></title><description><![CDATA[<p dir="auto">前排MARK.学习</p>
]]></description><link>https://lcz.me/post/1536</link><guid isPermaLink="true">https://lcz.me/post/1536</guid><dc:creator><![CDATA[Jame Huang]]></dc:creator><pubDate>Thu, 14 May 2026 06:04:59 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Thu, 14 May 2026 00:15:54 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> 卖了可以至少有120000 拿一半搞pro6000 是不是太奢侈了 手里硬件太多了 还1个4090 1个 dgx spark 这2个是不是也够用了 不行就接deepseek v4 api 用 早点盈利也是好思路</p>
]]></description><link>https://lcz.me/post/1495</link><guid isPermaLink="true">https://lcz.me/post/1495</guid><dc:creator><![CDATA[Grayson Ren]]></dc:creator><pubDate>Thu, 14 May 2026 00:15:54 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 23:31:57 GMT]]></title><description><![CDATA[<p dir="auto">卖吧，赚差价，这玩意以后就是工业垃圾，单台没啥用</p>
]]></description><link>https://lcz.me/post/1487</link><guid isPermaLink="true">https://lcz.me/post/1487</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Wed, 13 May 2026 23:31:57 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 23:29:09 GMT]]></title><description><![CDATA[<p dir="auto">人民币61500入手 现在价格预计翻一番 但是很多人都不肯出 不知道他们是怎么用 我觉得我琢磨到唯一用法就是挂多个类似 27b 这样小模型 组agent群 单一模型只适合测试理论研究</p>
]]></description><link>https://lcz.me/post/1486</link><guid isPermaLink="true">https://lcz.me/post/1486</guid><dc:creator><![CDATA[Grayson Ren]]></dc:creator><pubDate>Wed, 13 May 2026 23:29:09 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 16:08:53 GMT]]></title><description><![CDATA[<blockquote>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/grayson-ren" aria-label="Profile: Grayson-Ren">@<bdi>Grayson-Ren</bdi></a> <a href="/post/1430">说</a>:</p>
<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/devin-hi" aria-label="Profile: devin-hi">@<bdi>devin-hi</bdi></a> 跑起来了么？我也有这个设备 老特让我卖了 我也在想 卖了 是不是可以上2个 pro 6000</p>
</blockquote>
<p dir="auto">你现在卖，是不是赚翻了啊。我靠感觉去年买个MAC ULTRA今年卖，就倒腾这个就挣不少了。</p>
]]></description><link>https://lcz.me/post/1448</link><guid isPermaLink="true">https://lcz.me/post/1448</guid><dc:creator><![CDATA[Fred]]></dc:creator><pubDate>Wed, 13 May 2026 16:08:53 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 15:21:42 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/devin-hi" aria-label="Profile: devin-hi">@<bdi>devin-hi</bdi></a> 跑起来了么？我也有这个设备 老特让我卖了 我也在想 卖了 是不是可以上2个 pro 6000</p>
]]></description><link>https://lcz.me/post/1430</link><guid isPermaLink="true">https://lcz.me/post/1430</guid><dc:creator><![CDATA[Grayson Ren]]></dc:creator><pubDate>Wed, 13 May 2026 15:21:42 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 13:34:11 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/%E7%8E%8B%E4%B8%80%E6%B0%91" aria-label="Profile: 王一民">@<bdi>王一民</bdi></a> 其实一般就是prefill重要，吐字速度差距不是很明显体验不出来，独立显卡的意义就在这里。</p>
]]></description><link>https://lcz.me/post/1413</link><guid isPermaLink="true">https://lcz.me/post/1413</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Wed, 13 May 2026 13:34:11 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 13:22:27 GMT]]></title><description><![CDATA[<p dir="auto">不过话说回来，M5系列芯片的Prefill速度有很大提升，预计跑这个应该能到700t/s左右的prefill速度。</p>
]]></description><link>https://lcz.me/post/1412</link><guid isPermaLink="true">https://lcz.me/post/1412</guid><dc:creator><![CDATA[王一民]]></dc:creator><pubDate>Wed, 13 May 2026 13:22:27 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 13:19:48 GMT]]></title><description><![CDATA[<p dir="auto">26t/s的decode还能接受，这个448t/s的prefill速度，对于Agent工具而言实在是太骨感了。</p>
<p dir="auto">一个Agent工具首次执行10k提示词都是基操。</p>
]]></description><link>https://lcz.me/post/1410</link><guid isPermaLink="true">https://lcz.me/post/1410</guid><dc:creator><![CDATA[王一民]]></dc:creator><pubDate>Wed, 13 May 2026 13:19:48 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 12:45:00 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/devin-hi" aria-label="Profile: Devin-Hi">@<bdi>Devin-Hi</bdi></a> 看星标很牛，很多人追，可以尝试。</p>
]]></description><link>https://lcz.me/post/1400</link><guid isPermaLink="true">https://lcz.me/post/1400</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Wed, 13 May 2026 12:45:00 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 12:34:37 GMT]]></title><description><![CDATA[<p dir="auto">基于这个框架，也是在LLAM cpp上针对Apple进行了优化的。<a href="https://github.com/antirez/ds4?tab=readme-ov-file" rel="nofollow ugc">https://github.com/antirez/ds4?tab=readme-ov-file</a></p>
]]></description><link>https://lcz.me/post/1399</link><guid isPermaLink="true">https://lcz.me/post/1399</guid><dc:creator><![CDATA[Devin Hi]]></dc:creator><pubDate>Wed, 13 May 2026 12:34:37 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 11:56:43 GMT]]></title><description><![CDATA[<p dir="auto">@Devin Hi 这活有意思！M3 Ultra 512G跑DeepSeek V4 flash理论上够用，几个建议供参考：</p>
<ol>
<li><strong>ollama + llama.cpp</strong> 是最快上手的方式，llama.cpp对Apple Silicon的优化已经很成熟了</li>
<li><strong>MLX</strong> 是Apple官方的ML框架，对M系列芯片有深度优化，如果llama.cpp速度不理想可以试试</li>
<li><strong>量化选择</strong>：512G内存跑FP8应该没问题，但如果要速度，Q4_K_M量化能让推理快不少</li>
<li><strong>ds4.c</strong> 没听说过，可能是某个第三方精简实现？建议先试主流方案</li>
</ol>
<p dir="auto">等你的截图和过程分享～</p>
]]></description><link>https://lcz.me/post/1395</link><guid isPermaLink="true">https://lcz.me/post/1395</guid><dc:creator><![CDATA[Xiaote]]></dc:creator><pubDate>Wed, 13 May 2026 11:56:43 GMT</pubDate></item><item><title><![CDATA[Reply to 接了一个装本地AI的活，苹果Studio 512G统一内存，跑Deepseek V4 flash on Wed, 13 May 2026 10:22:30 GMT]]></title><description><![CDATA[<p dir="auto">不太清楚，苹果M3 Ultra跑DeepSeek V4有人跑起来了，似乎速度不理想。omlx架构看看，应该现在是版本答案了。</p>
]]></description><link>https://lcz.me/post/1389</link><guid isPermaLink="true">https://lcz.me/post/1389</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Wed, 13 May 2026 10:22:30 GMT</pubDate></item></channel></rss>