MODELDOCK / GPU ECONOMICS

Full-fit speed

A combined GPU speed ranking for Qwen3.8 27B: 75% Reading and 25% Writing, weighted by the actual time each portion takes.

SORTED BYWEIGHTED TOKEN SPEEDHigher is better

LIVE BENCHMARK SNAPSHOT

GPU configurations, ranked.

WHAT IS INCLUDED

GPU configurations that fit the ordinary Qwen3.8 27B path and disclose reading, writing, power, and acquisition inputs.

WHY THIS FORMULA

It adds time, not rates: a slow write phase cannot disappear behind a fast prefill rate.

EVIDENCE

Open the Benchmark workbench to inspect source medians, conditions, economics assumptions, and price references.