Arao AI
Back to search results

Qwen3-4B-Thinking-2507 · Q3_K_M

Start a discussion
AMD Radeon RX 9070 GRE12GBllama.cpp v0.35L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (Q3_K_M) on AMD Radeon RX 9070 GRE with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

100.99 tok/s

Decode

Decode speed

537.73 tok/s

Prefill

Prefill speed

0.18 s

TTFT

Time to first token

7.81 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 (current)Decode 100.99 · Prefill 537.73 ·
  2. llama.cpp · Q5_K_S Decode 85.62 · Prefill 376.76 ·

Core figures

Decode (avg)
100.99 tok/s
Prefill (avg)
537.73 tok/s
TTFT (avg)
0.18 s
VRAM (avg)
7.81 GB
MTP acceptance rate
—
TTFB
0.81 GB
Power draw
305.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B-Thinking-2507
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.35
Context length
2048 tokens
Batch size
512
GPU layers
28
Flash Attention
On

Hardware

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

GPU
AMD Radeon RX 9070 GRE
Nominal VRAM
12 GB
Measured VRAM (avg)
7.81 GB
OS
Ubuntu 24.04
Driver
amdgpu 6.10
CUDA
N/A
Power draw
305.6 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

    v0.35 · Ubuntu 24.04 · CUDA N/A · 2048 ctx

    100.99 tok/s

    Decode

    537.73 tok/s

    Prefill

    0.18 s

    TTFT

    7.81 GB

    VRAM

    —

    MTP

    305.6 W

    Power draw

    L2 batch seed 2026-10-13: llama.cpp on hw_id=32 running model_id=57 at Q3_K_M