Yes, we are back👑, with 206 tok/s on a single RTX 5090! Amazing Day-0 work from
Source
x.com
Author
Alibaba_Qwen
Date
Context
Running Qwen's Qwen3.8-27B on an RTX 5090 depends on the model representation and serving configuration. The launch announcement identifies NVFP4 plus DSpark as its optimized setup. SGLang's current deployment guide recommends NVFP4 for this class of GPU and lists approximately 16.5 GB for model weights. Additional memory must accommodate the running conversation and serving state, so fitting the weights is only part of making a local deployment work.
Terms in this piece · Glossary
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
