Benchmarking on Experiment, Fail, Learn, Repeat https://www.hairizuan.com/tags/benchmarking/ Recent content in Benchmarking on Experiment, Fail, Learn, Repeat Hugo -- gohugo.io en Copyright © 2026 Hairizuan Noorazman Mon, 02 Nov 2026 15:27:17 +0800 Don't Split a Model That Fits: One Pod vs Two for Gemma on CPU https://www.hairizuan.com/dont-split-a-model-that-fits-one-pod-vs-two-for-gemma-on-cpu/ Sun, 01 Nov 2026 00:00:00 +0000 https://www.hairizuan.com/dont-split-a-model-that-fits-one-pod-vs-two-for-gemma-on-cpu/ <p>In the <a href="https://www.hairizuan.com/serving-gemma-across-two-nodes-with-vllm-and-leaderworkerset-on-cpu/" >last post</a> I split a 1B-parameter Gemma model across two CPU nodes on purpose, to learn how <code>LeaderWorkerSet</code> and vLLM&rsquo;s multi-node mode behave. I said then that splitting a model that fits on one machine makes it slower, and I listed &ldquo;compare the split against a single replica&rdquo; among the things I had not done.</p>