Arao AI
Back to search results

Qwen3.8-9B · IQ2_M

Start a discussion
Apple M5 Ultra512GBllama.cpp v0.51L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (IQ2_M) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

130.42 tok/s

Decode

Decode speed

681.87 tok/s

Prefill

Prefill speed

2.23 s

TTFT

Time to first token

11.93 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · IQ2_M (current)Decode 130.42 · Prefill 681.87 ·
  2. Ollama · Q8_0 Decode 57.41 · Prefill 209.64 ·

Core figures

Decode (avg)
130.42 tok/s
Prefill (avg)
681.87 tok/s
TTFT (avg)
2.23 s
VRAM (avg)
11.93 GB
MTP acceptance rate
—
TTFB
1.04 GB
Power draw
131.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.51
Context length
2048 tokens
Batch size
256
GPU layers
33
Flash Attention
Off

Hardware

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

GPU
Apple M5 Ultra
Nominal VRAM
512 GB
Measured VRAM (avg)
11.93 GB
OS
macOS 26.1
Power draw
131.7 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.51 · macOS 26.1 · 2048 ctx

    130.42 tok/s

    Decode

    681.87 tok/s

    Prefill

    2.23 s

    TTFT

    11.93 GB

    VRAM

    —

    MTP

    131.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=29 running model_id=52 at IQ2_M