<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>MPS on 111qqz's blog</title><link>https://111qqz.com/en/tags/mps/</link><description>Recent content in MPS on 111qqz's blog</description><generator>Hugo -- gohugo.io</generator><language>en</language><managingEditor>hust.111qqz@gmail.com (111qqz)</managingEditor><webMaster>hust.111qqz@gmail.com (111qqz)</webMaster><copyright>© 2011-2026 111qqz</copyright><lastBuildDate>Wed, 18 Nov 2026 18:40:00 +0800</lastBuildDate><atom:link href="https://111qqz.com/en/tags/mps/index.xml" rel="self" type="application/rss+xml"/><item><title>NVIDIA MPS：既然有 CUDA Streams，为什么还需要跨进程并发？</title><link>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-11-18-nvidia-mps-cuda-context-stream-concurrency/</link><pubDate>Wed, 18 Nov 2026 18:40:00 +0800</pubDate><author>hust.111qqz@gmail.com (111qqz)</author><guid>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-11-18-nvidia-mps-cuda-context-stream-concurrency/</guid><description>&lt;p&gt;之前在写 &lt;a href="https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-11-14-gpu-execution-model-latency-hiding/" &gt;GPU Execution Model&lt;/a&gt; 的时候，我对 SM、Warp、Block 这些概念已经有了基本的理解。但最近碰到一个场景让我停下来想了一会儿：我们有两个独立的推理服务共享同一张 A100，各自的 GPU 利用率都不高，直觉上 GPU 应该有很多空闲的执行资源。&lt;/p&gt;</description></item></channel></rss>