Qwen3.8 27B

on NVIDIA RTX 5090 · llama.cpp · 90,112 ctx

Sep 7, 2026

Throughput

97.7 t/s gen · 1995.8 tokens/s prompt processing (prefill)Prompt processing measures input prefill speed; it does not tell you the time to first token.

Quant

Q8_K_L (GGUF)

KV cache

VRAM reported

32 GB

Summary

User benchmarks DFlash2 speculative decoding against MTP on Qwen3.8-27B. DFlash2 gives 20% faster generation but reduces context size by 38%. The prompt and generation TPS figures are weighted averages.

View source ↗