Arao AI
Back to search results

gemma-4-12B-it · IQ3_S

Start a discussion
Apple M5 Max128GBllama.cpp v0.30L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-12B-it (IQ3_S) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

67.49 tok/s

Decode

Decode speed

263.38 tok/s

Prefill

Prefill speed

2.02 s

TTFT

Time to first token

18.73 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. llama.cpp · IQ3_S (current)Decode 67.49 · Prefill 263.38 ·
  2. Ollama · IQ2_M Decode 66.82 · Prefill 219.12 ·
  3. Ollama · Q5_K_M Decode 43.37 · Prefill 143.55 ·
  4. Strata · Q4_K_S Decode 224.6 · Prefill 779.88 ·

Core figures

Decode (avg)
67.49 tok/s
Prefill (avg)
263.38 tok/s
TTFT (avg)
2.02 s
VRAM (avg)
18.73 GB
MTP acceptance rate
—
TTFB
0.54 GB
Power draw
55.7 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.30
Context length
4096 tokens
Batch size
512
GPU layers
99
Flash Attention
On

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)
18.73 GB
OS
macOS 26.1
Power draw
55.7 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

    v0.30 · macOS 26.1 · 4096 ctx

    67.49 tok/s

    Decode

    263.38 tok/s

    Prefill

    2.02 s

    TTFT

    18.73 GB

    VRAM

    —

    MTP

    55.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=65 at IQ3_S