Arao AI
Back to search results

gemma-4-31B-it · IQ2_XS

Start a discussion
Apple M5 Max128GBllama.cpp v0.53L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-31B-it (IQ2_XS) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

58.24 tok/s

Decode

Decode speed

303.37 tok/s

Prefill

Prefill speed

2.49 s

TTFT

Time to first token

33.35 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 · IQ1_S Decode 64.71 · Prefill 267.16 ·
  2. llama.cpp · IQ2_XS (current)Decode 58.24 · Prefill 303.37 ·
  3. Ollama · Q4_K_M Decode 32.9 · Prefill 145.14 ·
  4. Strata · IQ4_XS Decode 138.4 · Prefill 718.68 ·
  5. Strata · Q3_K_M Decode 149.39 · Prefill 771.31 ·

Core figures

Decode (avg)
58.24 tok/s
Prefill (avg)
303.37 tok/s
TTFT (avg)
2.49 s
VRAM (avg)
33.35 GB
MTP acceptance rate
—
TTFB
0.47 GB
Power draw
40.8 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-31B-it
Quantization
IQ2_XS
Framework
llama.cpp
Version
v0.53
Context length
8192 tokens
Batch size
512
GPU layers
28
Flash Attention
Off

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)
33.35 GB
OS
macOS 26.1
Power draw
40.8 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

    v0.53 · macOS 26.1 · 8192 ctx

    58.24 tok/s

    Decode

    303.37 tok/s

    Prefill

    2.49 s

    TTFT

    33.35 GB

    VRAM

    —

    MTP

    40.8 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=63 at IQ2_XS