This is what it looks like to run qwen 3.8 27B locally on a gaming pc
Around 115 t/s
I didn’t know it before, until I used DGX Spark to test Qwen 3.8-27B and Ling-3.0-flash in the past few days.
I came to the conclusion that Spark’s memory bandwidth is a flaw😅 1. It is destined that large models can run, but they cannot run fast. 2. The small model can run, but the graphics card cannot run it. Screen recording: Qwen 3.8 27B multi-concurrency test, maximum 32 tok/s 🙃…
Qwen 3.8 27B vs GLM 5.3 vs Kimi K3 vs Opus 5
You have reached the end of the archive
All of qwen38