Arao AI
Back to search results

Qwen3.5-27B · IQ4_XS

Start a discussion
NVIDIA RTX 509032GBvLLM 0.7.3L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (IQ4_XS) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

108 tok/s

Decode

Decode speed

1,288 tok/s

Prefill

Prefill speed

0.71 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 134 · Prefill 1478 ·
  2. llama.cpp · Q5_K_M Decode 114 · Prefill 1331 ·
  3. llama.cpp · Q8_0 Decode 92 · Prefill 792 ·
  4. vLLM · IQ4_XS (current)Decode 108 · Prefill 1288 ·
  5. vLLM · Q4_K_M Decode 100 · Prefill 1182 ·
  6. vLLM · Q5_K_M Decode 106 · Prefill 956 ·
  7. vLLM · Q8_0 Decode 74 · Prefill 795 ·
  8. Ollama · IQ4_XS Decode 136 · Prefill 1332 ·
  9. Ollama · Q4_K_M Decode 125 · Prefill 1177 ·
  10. Ollama · Q5_K_M Decode 113 · Prefill 1178 ·
  11. Ollama · Q8_0 Decode 75 · Prefill 899 ·
  12. Strata · IQ4_XS Decode 198 · Prefill 1923 ·
  13. Strata · Q4_K_M Decode 178 · Prefill 1794 ·
  14. Strata · Q5_K_M Decode 147 · Prefill 1764 ·
  15. Strata · Q8_0 Decode 127 · Prefill 1290 ·
  16. llama.cpp · Q4_K_M Decode 30.09 · Prefill 1515.83 ·

Core figures

Decode (avg)
108 tok/s
Prefill (avg)
1,288 tok/s
TTFT (avg)
0.71 s
VRAM (avg)
— GB
MTP acceptance rate
—
TTFB
— GB
Power draw
499 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 5090
Nominal VRAM
32 GB
OS
Linux 6.8.0
Driver
570.86.15
CUDA
12.6
Power draw
499 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

    108 tok/s

    Decode

    1,288 tok/s

    Prefill

    0.71 s

    TTFT

    0 GB

    VRAM

    —

    MTP

    499 W

    Power draw

    Auto-generated LLM-sourced record 2026-10-09. Model: Qwen3.5-27B, Hardware: RTX 5090, Framework: 2, Quant: IQ4_XS.