BLAST RADIUS

Open-weight & owned stack · Quantization efficiency

Qwen3.8-27B Achieves High Throughput On Two GPUs

Sep 22, 2026

A user reported serving an INT4 Qwen3.8-27B model on two RTX 3090s with full 262k context and high decode and prefill throughput.

Read the original at reddit.comOpens the publisher's site in a new tab

More in Open-weight & owned stack