Arao AI
Back to search results

Qwen3.8-9B · IQ3_S

Start a discussion
Apple M3 Max36GBllama.cpp v0.42L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (IQ3_S) on Apple M3 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

126.59 tok/s

Decode

Decode speed

664.89 tok/s

Prefill

Prefill speed

1.15 s

TTFT

Time to first token

14.57 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ2_M Decode 136.47 · Prefill 438.69 ·
  2. llama.cpp · IQ3_S (current)Decode 126.59 · Prefill 664.89 ·
  3. Strata · Q2_K Decode 180.32 · Prefill 900.55 ·

Core figures

Decode (avg)
126.59 tok/s
Prefill (avg)
664.89 tok/s
TTFT (avg)
1.15 s
VRAM (avg)
14.57 GB
MTP acceptance rate
—
TTFB
1.01 GB
Power draw
53.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.42
Context length
2048 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
Apple M3 Max
Nominal VRAM
36 GB
Measured VRAM (avg)
14.57 GB
OS
macOS 15.0
Driver
Apple M3 Max
CUDA
N/A
Power draw
53.6 W

Sources & evidence

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

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

    v0.42 · macOS 15.0 · CUDA N/A · 2048 ctx

    126.59 tok/s

    Decode

    664.89 tok/s

    Prefill

    1.15 s

    TTFT

    14.57 GB

    VRAM

    —

    MTP

    53.6 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=4 running model_id=52 at IQ3_S