<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>MegaMoE on TechLLM</title><link>https://techllm.github.io/tags/megamoe/</link><description>Recent content in MegaMoE on TechLLM</description><generator>Hugo</generator><language>ko</language><lastBuildDate>Wed, 10 Jun 2026 23:01:35 +0900</lastBuildDate><atom:link href="https://techllm.github.io/tags/megamoe/index.xml" rel="self" type="application/rss+xml"/><item><title>DeepSeek V4 출시 43일, 추론 최적화 전쟁의 실제 성적표</title><link>https://techllm.github.io/posts/deepseek-v4-43day-inference-optimization-race/</link><pubDate>Wed, 10 Jun 2026 23:01:35 +0900</pubDate><guid>https://techllm.github.io/posts/deepseek-v4-43day-inference-optimization-race/</guid><description>DeepSeek V4 1.6T 공개 이후 43일 동안 NVIDIA, AMD, 화웨이 진영이 어떻게 성능을 끌어올렸는지 — SemiAnalysis 분석을 정리했습니다. AMD MI355X의 100배 향상, MegaMoE 커널, CSA/HCA, 그리고 NVIDIA TensorRT-LLM의 출시 첫날 사고까지.</description></item></channel></rss>