Arao AI
Back to search results

Qwen3.8-9B · Q3_K_L

Start a discussion
Apple M5 Max128GBvLLM v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (Q3_K_L) on Apple M5 Max with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

202.52 tok/s

Decode

Decode speed

1,021.73 tok/s

Prefill

Prefill speed

2.01 s

TTFT

Time to first token

15.71 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. vLLM · Q3_K_L (current)Decode 202.52 · Prefill 1021.73 ·
  2. Strata · Q3_K_M Decode 280.05 · Prefill 1450.84 ·
  3. Strata · IQ3_XS Decode 320.14 · Prefill 1058.22 ·
  4. Strata · Q2_K Decode 331.24 · Prefill 1237.62 ·

Core figures

Decode (avg)
202.52 tok/s
Prefill (avg)
1,021.73 tok/s
TTFT (avg)
2.01 s
VRAM (avg)
15.71 GB
MTP acceptance rate
—
TTFB
1.62 GB
Power draw
212.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
Q3_K_L
Framework
vLLM
Version
v0.52
Context length
32768 tokens
Batch size
16
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
Apple M5 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
15.71 GB
OS
Ubuntu 22.04
Driver
555.85
CUDA
12.8
Power draw
212.6 W

Sources & evidence

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

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

    v0.52 · Ubuntu 22.04 · CUDA 12.8 · 32768 ctx

    202.52 tok/s

    Decode

    1,021.73 tok/s

    Prefill

    2.01 s

    TTFT

    15.71 GB

    VRAM

    —

    MTP

    212.6 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=28 running model_id=52 at Q3_K_L