Arao AI
Back to search results

Qwen3.5-4B · Q5_K_S

Start a discussion
NVIDIA RTX 309024GBllama.cpp v0.45L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q5_K_S) on NVIDIA RTX 3090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

139.06 tok/s

Decode

Decode speed

548.16 tok/s

Prefill

Prefill speed

2.37 s

TTFT

Time to first token

9.66 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q5_K_S (current)Decode 139.06 · Prefill 548.16 ·
  2. llama.cpp · Q6_K Decode 108.41 · Prefill 420.36 ·
  3. vLLM · Q5_K_S Decode 129.61 · Prefill 696.02 ·

Core figures

Decode (avg)
139.06 tok/s
Prefill (avg)
548.16 tok/s
TTFT (avg)
2.37 s
VRAM (avg)
9.66 GB
MTP acceptance rate
—
TTFB
1.11 GB
Power draw
470.5 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q5_K_S
Framework
llama.cpp
Version
v0.45
Context length
2048 tokens
Batch size
256
GPU layers
20
Flash Attention
Off

Hardware

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

GPU
NVIDIA RTX 3090
Nominal VRAM
24 GB
Measured VRAM (avg)
9.66 GB
OS
Windows 11 23H2
Driver
555.85
CUDA
12.4
Power draw
470.5 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

    v0.45 · Windows 11 23H2 · CUDA 12.4 · 2048 ctx

    139.06 tok/s

    Decode

    548.16 tok/s

    Prefill

    2.37 s

    TTFT

    9.66 GB

    VRAM

    —

    MTP

    470.5 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=2 running model_id=48 at Q5_K_S