https://k4i.top/zh/tags/distributed/Recent content in Distributed on k4i's blogHugo -- gohugo.iozhsky_io@outlook.com (K4i)sky_io@outlook.com (K4i)All content is subject to the license of <a rel="license noopener" href="https://creativecommons.org/licenses/by-nc-sa/4.0/" target="_blank">CC BY-NC-SA 4.0</a> .Wed, 22 Apr 2026 12:00:00 +0800https://k4i.top/zh/posts/disaggregated-prefill/Wed, 22 Apr 2026 12:00:00 +0800sky_io@outlook.com (K4i)Sat, 30 May 2026 00:04:33 +0800https://k4i.top/zh/posts/disaggregated-prefill/<h2 id="ceiling">为什么同一张 GPU 上的共存有上限</h2> <p><a href="https://k4i.top/zh/posts/chunked-prefill/">chunked prefill</a> 通过把 prefill 切成小块，让 prefill 和 decode 在同一张 GPU 上共存得更平滑。但即使 chunk 切得再好，prefill 和 decode 仍然在<em>共享同一张 GPU</em>。它们会竞争：</p>K4ifeatured imagellminferencesystemsdistributedschedulingAILLM Inference Internals