---
title: "I have made an update to my theoretical upper bound calculation to also predict prefill speed"
date: 2026-08-09
canonical: https://solmaz.io/x/2086346067952677215/
x_url: https://x.com/onusoz/status/2086346067952677215
license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
---

I have made an update to my theoretical upper bound calculation to also predict prefill speed

Prefill relaxes the assumption we make for decode, that it is only be memory bottlenecked. So prefill can be both compute or memory bottlenecked. I use the FLOP limits reported by hardware producers for the estimates:

These estimates will also be available in https://ourmodels.cc/ for indexed model and hardware in a couple days, once a long running job finishes

Blog post: https://solmaz.io/llm-performance-upper-bounds