<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:sy="http://purl.org/rss/1.0/modules/syndication/" xmlns:media="http://search.yahoo.com/mrss/"><channel><title>Chip on Shaowen Chen's Website</title><link>https://www.chenshaowen.com/en/tags/chip/</link><description>Recent content in Chip on Shaowen Chen's Website</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>&amp;copy;2016 - {year}, All Rights Reserved.</copyright><lastBuildDate>Tue, 19 Mar 2024 00:00:00 +0000</lastBuildDate><sy:updatePeriod>weekly</sy:updatePeriod><atom:link href="https://www.chenshaowen.com/en/tags/chip/atom.xml" rel="self" type="application/rss+xml"/><item><title>AI Chip High-Speed Interconnect Solutions</title><link>https://www.chenshaowen.com/en/blog/ai-chip-high-speed-connection-solution.html</link><pubDate>Tue, 19 Mar 2024 00:00:00 +0000</pubDate><atom:modified>Tue, 19 Mar 2024 00:00:00 +0000</atom:modified><guid>https://www.chenshaowen.com/en/blog/ai-chip-high-speed-connection-solution.html</guid><description>I have been studying the infrastructure around model training lately, and I found that AI chip interconnect topology determines the scheduling and resource allocation of training cluster jobs. So I spent a little time putting together the common AI chip interconnect solutions.
1. Point-to-Point Interconnect In a traditional PCIe system, data transfer between AI chips goes over PCIe, which cannot meet the demands of large-scale data transfer.</description><dc:creator>微信公众号</dc:creator><category>AI</category><category>Chip</category><category>Interconnect</category><category>Hardware</category><category>Operations</category></item></channel></rss>