Arao AI
Back to search results

Qwen3.5-4B · IQ3_XS

Start a discussion
AMD Radeon AI PRO R970032GBvLLM v0.46L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ3_XS) on AMD Radeon AI PRO R9700 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

208.55 tok/s

Decode

Decode speed

738.48 tok/s

Prefill

Prefill speed

0.39 s

TTFT

Time to first token

7.31 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_M Decode 183.89 · Prefill 727.67 ·
  2. vLLM · IQ3_XS (current)Decode 208.55 · Prefill 738.48 ·
  3. vLLM · Q5_K_M Decode 122.86 · Prefill 611.36 ·

Core figures

Decode (avg)
208.55 tok/s
Prefill (avg)
738.48 tok/s
TTFT (avg)
0.39 s
VRAM (avg)
7.31 GB
MTP acceptance rate
—
TTFB
1.67 GB
Power draw
317.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ3_XS
Framework
vLLM
Version
v0.46
Context length
32768 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
AMD Radeon AI PRO R9700
Nominal VRAM
32 GB
Measured VRAM (avg)
7.31 GB
OS
Ubuntu 24.04
Driver
amdgpu 6.10
CUDA
N/A
Power draw
317.3 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.46 · Ubuntu 24.04 · CUDA N/A · 32768 ctx

    208.55 tok/s

    Decode

    738.48 tok/s

    Prefill

    0.39 s

    TTFT

    7.31 GB

    VRAM

    —

    MTP

    317.3 W

    Power draw

    L2 batch seed 2026-10-13: vLLM on hw_id=79 running model_id=48 at IQ3_XS