Arao AI
Back to search results

gemma-4-E4B-it · Q3_K_S

Start a discussion
Apple M2 Ultrallama.cpp v0.29L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-E4B-it (Q3_K_S) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

151.33 tok/s

Decode

Decode speed

747.84 tok/s

Prefill

Prefill speed

1.57 s

TTFT

Time to first token

7.56 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_S (current)Decode 151.33 · Prefill 747.84 ·
  2. llama.cpp · Q5_K_S Decode 114.93 · Prefill 613.44 ·
  3. vLLM · Q5_K_S Decode 123.25 · Prefill 491.18 ·

Core figures

Decode (avg)
151.33 tok/s
Prefill (avg)
747.84 tok/s
TTFT (avg)
1.57 s
VRAM (avg)
7.56 GB
MTP acceptance rate
—
TTFB
1.21 GB
Power draw
40.3 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-E4B-it
Quantization
Q3_K_S
Framework
llama.cpp
Version
v0.29
Context length
4096 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 M2 Ultra
Measured VRAM (avg)
7.56 GB
OS
macOS 15.0
Driver
Apple M2 Ultra
CUDA
N/A
Power draw
40.3 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

    v0.29 · macOS 15.0 · CUDA N/A · 4096 ctx

    151.33 tok/s

    Decode

    747.84 tok/s

    Prefill

    1.57 s

    TTFT

    7.56 GB

    VRAM

    —

    MTP

    40.3 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=103 running model_id=67 at Q3_K_S