Arao AI
Back to search results

Qwen3.5-4B · Q3_K_S

Start a discussion
NVIDIA RTX 509032GBvLLM v0.53L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q3_K_S) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

324.89 tok/s

Decode

Decode speed

1,253.12 tok/s

Prefill

Prefill speed

1.45 s

TTFT

Time to first token

7.56 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q3_K_L Decode 338.21 · Prefill 1409.78 ·
  2. vLLM · Q3_K_M Decode 356.1 · Prefill 1633.97 ·
  3. vLLM · Q3_K_S (current)Decode 324.89 · Prefill 1253.12 ·
  4. Strata · Q4_K_S Decode 353.67 · Prefill 1676.75 ·

Core figures

Decode (avg)
324.89 tok/s
Prefill (avg)
1,253.12 tok/s
TTFT (avg)
1.45 s
VRAM (avg)
7.56 GB
MTP acceptance rate
—
TTFB
2.6 GB
Power draw
449 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q3_K_S
Framework
vLLM
Version
v0.53
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 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
7.56 GB
OS
Ubuntu 24.04
Driver
560.94
CUDA
12.8
Power draw
449 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

    v0.53 · Ubuntu 24.04 · CUDA 12.8 · 4096 ctx

    324.89 tok/s

    Decode

    1,253.12 tok/s

    Prefill

    1.45 s

    TTFT

    7.56 GB

    VRAM

    —

    MTP

    449 W

    Power draw

    L2 batch seed 2026-10-10: vLLM on hw_id=13 running model_id=48 at Q3_K_S