Single RTX 5090 Runs Qwen 27B at 200+ t/s with ninfer

A Reddit user ran the NVFP4-quantized Qwen3.8-27B model on a single RTX 5090 (32GB), comparing vLLM with the new ninfer engine. Tests showed ninfer achieving over 200 tokens/s decoding speed.

2026-08-31 ~ 2026-08-31 · 2 related posts

• Running Qwen 27B on a Single 5090: 200+ t/s Inference Benchmarks — Maleficent-Ad5999 · 2026-08-31
• vLLM vs ninfer for Qwen 3.8-27B on RTX 5090: a Reddit reality check — MaxKingCS · 2026-08-31