Arao AI
Back to search results

Qwen3.8-9B · Q5_K_M

Start a discussion
AMD Ryzen AI Max+ 395128GBvLLM v0.32L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (Q5_K_M) on AMD Ryzen AI Max+ 395 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

82.89 tok/s

Decode

Decode speed

274.27 tok/s

Prefill

Prefill speed

1.87 s

TTFT

Time to first token

19.68 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 · IQ2_M Decode 125.51 · Prefill 634.19 ·
  2. llama.cpp · IQ3_S Decode 115.85 · Prefill 489.76 ·
  3. vLLM · Q5_K_M (current)Decode 82.89 · Prefill 274.27 ·
  4. vLLM · IQ2_XS Decode 145.04 · Prefill 523.91 ·
  5. vLLM · Q3_K_L Decode 107.95 · Prefill 386.71 ·
  6. Strata · IQ3_XS Decode 164.91 · Prefill 583.87 ·
  7. Strata · Q5_K_S Decode 114.19 · Prefill 509.48 ·

Core figures

Decode (avg)
82.89 tok/s
Prefill (avg)
274.27 tok/s
TTFT (avg)
1.87 s
VRAM (avg)
19.68 GB
MTP acceptance rate
—
TTFB
0.66 GB
Power draw
111.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
Q5_K_M
Framework
vLLM
Version
v0.32
Context length
4096 tokens
Batch size
8
GPU layers
999
Flash Attention
Off

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)
19.68 GB
OS
macOS 15.0
Driver
570.86
CUDA
N/A
Power draw
111.8 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

    v0.32 · macOS 15.0 · CUDA N/A · 4096 ctx

    82.89 tok/s

    Decode

    274.27 tok/s

    Prefill

    1.87 s

    TTFT

    19.68 GB

    VRAM

    —

    MTP

    111.8 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=24 running model_id=52 at Q5_K_M