Arao AI
Back to search results

Qwen3-8B · IQ2_M

Start a discussion
Apple M3 Max36GBllama.cpp v0.20L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (IQ2_M) on Apple M3 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

137.92 tok/s

Decode

Decode speed

603.9 tok/s

Prefill

Prefill speed

1.57 s

TTFT

Time to first token

10.84 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · FP16 Decode 58.36 · Prefill 290.84 ·
  2. llama.cpp · IQ2_M (current)Decode 137.92 · Prefill 603.9 ·
  3. vLLM · FP16 Decode 57.8 · Prefill 255.66 ·
  4. vLLM · IQ2_M Decode 145.31 · Prefill 613.3 ·
  5. Ollama · IQ2_M Decode 129.83 · Prefill 636.38 ·

Core figures

Decode (avg)
137.92 tok/s
Prefill (avg)
603.9 tok/s
TTFT (avg)
1.57 s
VRAM (avg)
10.84 GB
MTP acceptance rate
—
TTFB
1.1 GB
Power draw
73.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.20
Context length
4096 tokens
Batch size
512
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 M3 Max
Nominal VRAM
36 GB
Measured VRAM (avg)
10.84 GB
OS
macOS 15.0
Driver
Apple M3 Max
CUDA
N/A
Power draw
73.7 W

Sources & evidence

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

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

    v0.20 · macOS 15.0 · CUDA N/A · 4096 ctx

    137.92 tok/s

    Decode

    603.9 tok/s

    Prefill

    1.57 s

    TTFT

    10.84 GB

    VRAM

    —

    MTP

    73.7 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=4 running model_id=53 at IQ2_M