Arao AI
Back to search results

Qwen3.5-27B · IQ4_XS

Start a discussion
NVIDIA RTX 4090 D24GBvLLM 0.7.3L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (IQ4_XS) on NVIDIA RTX 4090 D with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

78 tok/s

Decode

Decode speed

936 tok/s

Prefill

Prefill speed

0.99 s

TTFT

Time to first token

— GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ4_XS Decode 103 · Prefill 1079 ·
  2. llama.cpp · Q4_K_M Decode 92 · Prefill 992 ·
  3. llama.cpp · Q5_K_M Decode 82 · Prefill 977 ·
  4. llama.cpp · Q8_0 Decode 69 · Prefill 646 ·
  5. vLLM · IQ4_XS (current)Decode 78 · Prefill 936 ·
  6. vLLM · Q4_K_M Decode 72 · Prefill 862 ·
  7. vLLM · Q5_K_M Decode 79 · Prefill 770 ·
  8. vLLM · Q8_0 Decode 57 · Prefill 577 ·
  9. Ollama · IQ4_XS Decode 101 · Prefill 1055 ·
  10. Ollama · Q4_K_M Decode 92 · Prefill 930 ·
  11. Ollama · Q5_K_M Decode 87 · Prefill 850 ·
  12. Ollama · Q8_0 Decode 54 · Prefill 647 ·
  13. Strata · IQ4_XS Decode 151 · Prefill 1357 ·
  14. Strata · Q4_K_M Decode 136 · Prefill 1276 ·
  15. Strata · Q5_K_M Decode 110 · Prefill 1311 ·
  16. Strata · Q8_0 Decode 93 · Prefill 1001 ·

Core figures

Decode (avg)
78 tok/s
Prefill (avg)
936 tok/s
TTFT (avg)
0.99 s
VRAM (avg)
— GB
MTP acceptance rate
—
TTFB
— GB
Power draw
492 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-27B
Quantization
IQ4_XS
Framework
vLLM
Version
0.7.3
Context length
8192 tokens
Batch size
512
GPU layers
99
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 D
Nominal VRAM
24 GB
OS
Linux 6.8.0
Driver
570.86.15
CUDA
12.6
Power draw
492 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-09

    0.7.3 · Linux 6.8.0 · CUDA 12.6 · 8192 ctx

    78 tok/s

    Decode

    936 tok/s

    Prefill

    0.99 s

    TTFT

    0 GB

    VRAM

    —

    MTP

    492 W

    Power draw

    Auto-generated LLM-sourced record 2026-10-09. Model: Qwen3.5-27B, Hardware: RTX 4090 D, Framework: 2, Quant: IQ4_XS.