Arao AI
Back to search results

Qwen3.5-27B · IQ1_S

Start a discussion
NVIDIA RTX 409024GBvLLM v0.41L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (IQ1_S) on NVIDIA RTX 4090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

153.29 tok/s

Decode

Decode speed

579.99 tok/s

Prefill

Prefill speed

1.19 s

TTFT

Time to first token

22.8 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. vLLM · IQ1_S (current)Decode 153.29 · Prefill 579.99 ·
  2. llama.cpp · IQ4_XS Decode 100 · Prefill 1111 ·
  3. llama.cpp · Q4_K_M Decode 89 · Prefill 1010 ·
  4. llama.cpp · Q5_K_M Decode 86 · Prefill 1000 ·
  5. llama.cpp · Q8_0 Decode 69 · Prefill 586 ·
  6. vLLM · IQ4_XS Decode 81 · Prefill 964 ·
  7. vLLM · Q4_K_M Decode 75 · Prefill 884 ·
  8. vLLM · Q5_K_M Decode 80 · Prefill 712 ·
  9. vLLM · Q8_0 Decode 55 · Prefill 596 ·
  10. Ollama · IQ4_XS Decode 102 · Prefill 988 ·
  11. Ollama · Q4_K_M Decode 94 · Prefill 874 ·
  12. Ollama · Q5_K_M Decode 85 · Prefill 894 ·
  13. Ollama · Q8_0 Decode 57 · Prefill 682 ·
  14. Strata · IQ4_XS Decode 148 · Prefill 1453 ·
  15. Strata · Q4_K_M Decode 133 · Prefill 1354 ·
  16. Strata · Q5_K_M Decode 111 · Prefill 1332 ·
  17. Strata · Q8_0 Decode 95 · Prefill 955 ·

Core figures

Decode (avg)
153.29 tok/s
Prefill (avg)
579.99 tok/s
TTFT (avg)
1.19 s
VRAM (avg)
22.8 GB
MTP acceptance rate
—
TTFB
1.23 GB
Power draw
437.2 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-27B
Quantization
IQ1_S
Framework
vLLM
Version
v0.41
Context length
4096 tokens
Batch size
8
GPU layers
999
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
NVIDIA RTX 4090
Nominal VRAM
24 GB
Measured VRAM (avg)
22.8 GB
OS
Windows 11 23H2
Driver
565.57
CUDA
12.6
Power draw
437.2 W

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-10

    v0.41 · Windows 11 23H2 · CUDA 12.6 · 4096 ctx

    153.29 tok/s

    Decode

    579.99 tok/s

    Prefill

    1.19 s

    TTFT

    22.8 GB

    VRAM

    —

    MTP

    437.2 W

    Power draw

    L2 batch seed 2026-10-10: vLLM on hw_id=1 running model_id=50 at IQ1_S