Arao AI
Back to search results

Llama-3.2-3B-Instruct · Q5_K_M

Start a discussion
Apple M2 Ultrallama.cpp v0.30L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (Q5_K_M) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

233.78 tok/s

Decode

Decode speed

783.42 tok/s

Prefill

Prefill speed

2.4 s

TTFT

Time to first token

7.96 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 · Q5_K_M (current)Decode 233.78 · Prefill 783.42 ·
  2. Ollama · Q5_K_M Decode 207.01 · Prefill 946.05 ·

Core figures

Decode (avg)
233.78 tok/s
Prefill (avg)
783.42 tok/s
TTFT (avg)
2.4 s
VRAM (avg)
7.96 GB
MTP acceptance rate
—
TTFB
1.87 GB
Power draw
110 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.2-3B-Instruct
Quantization
Q5_K_M
Framework
llama.cpp
Version
v0.30
Context length
16384 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 M2 Ultra
Measured VRAM (avg)
7.96 GB
OS
macOS 26.1
Power draw
110 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 · 16384 ctx

    233.78 tok/s

    Decode

    783.42 tok/s

    Prefill

    2.4 s

    TTFT

    7.96 GB

    VRAM

    —

    MTP

    110 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=103 running model_id=70 at Q5_K_M