Arao AI
Back to search results

Qwen3-1.7B · Q4_K_M

Start a discussion
AMD Ryzen AI Max+ 395128GBllama.cpp v0.51L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-1.7B (Q4_K_M) on AMD Ryzen AI Max+ 395 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

187.84 tok/s

Decode

Decode speed

854.19 tok/s

Prefill

Prefill speed

0.33 s

TTFT

Time to first token

5.03 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · Q4_K_M (current)Decode 187.84 · Prefill 854.19 ·
  2. vLLM · IQ1_S Decode 304.08 · Prefill 1413.06 ·
  3. llama.cpp · Q5_K_M Decode 145.14 · Prefill 763.75 ·
  4. vLLM · IQ3_S Decode 246.99 · Prefill 860.32 ·
  5. vLLM · Q8_0 Decode 132.62 · Prefill 633.28 ·

Core figures

Decode (avg)
187.84 tok/s
Prefill (avg)
854.19 tok/s
TTFT (avg)
0.33 s
VRAM (avg)
5.03 GB
MTP acceptance rate
—
TTFB
1.5 GB
Power draw
114.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-1.7B
Quantization
Q4_K_M
Framework
llama.cpp
Version
v0.51
Context length
32768 tokens
Batch size
512
GPU layers
20
Flash Attention
On

Hardware

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

GPU
AMD Ryzen AI Max+ 395
Nominal VRAM
128 GB
Measured VRAM (avg)
5.03 GB
OS
macOS 15.0
Driver
560.94
CUDA
N/A
Power draw
114.8 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.51 · macOS 15.0 · CUDA N/A · 32768 ctx

    187.84 tok/s

    Decode

    854.19 tok/s

    Prefill

    0.33 s

    TTFT

    5.03 GB

    VRAM

    —

    MTP

    114.8 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=24 running model_id=55 at Q4_K_M