TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Running DeepSeek V3 671B on M4 Mac Mini Cluster

8 点作者 chadash5 个月前

2 条评论

jauntywundrkind5 个月前
Shout out to the video team, for this super cute tower of minis next to the Christmas tree.<p><a href="https:&#x2F;&#x2F;x.com&#x2F;exolabs&#x2F;status&#x2F;1872444906851229814" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;exolabs&#x2F;status&#x2F;1872444906851229814</a><p>Only just considering now that Strix Halo could help fill this gap that Mac chips with their huge memory bandwidth enjoy. 256GB systems shouldn&#x27;t be hard to build!!<p>MI300a APU seems not popular but for consumers, this mix of big CPU and GPU seems perhaps quite compelling!
talldayo5 个月前
&gt; The M4 Max has 546GB&#x2F;s of memory bandwidth and ~34TFLOPS (fp16) = ~68 GB&#x2F;s, a ratio of ~8.02. Whereas NVIDIA RTX 4090 has 1008GB&#x2F;s memory bandwidth and ~330TFLOPS (fp16) = ~660GB&#x2F;s, a ratio of ~1.52.<p>Why are we comparing FP16 performance when you&#x27;re inferencing INT4 quantized models? Seems like a misleading figure to compare with when it&#x27;s not really even the performance you&#x27;re measuring.
评论 #42523863 未加载