Arao AI
Back to search results

Llama-3.1-8B-Instruct · Q4_K_S

Start a discussion
Apple M2 Ultrallama.cpp v0.34L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.1-8B-Instruct (Q4_K_S) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

123.15 tok/s

Decode

Decode speed

540.31 tok/s

Prefill

Prefill speed

2.07 s

TTFT

Time to first token

15.2 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 · Q3_K_L Decode 129.48 · Prefill 476.49 ·
  2. llama.cpp · Q4_K_S (current)Decode 123.15 · Prefill 540.31 ·
  3. Ollama · Q4_K_S Decode 103.18 · Prefill 542.74 ·

Core figures

Decode (avg)
123.15 tok/s
Prefill (avg)
540.31 tok/s
TTFT (avg)
2.07 s
VRAM (avg)
15.2 GB
MTP acceptance rate
—
TTFB
0.99 GB
Power draw
142.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.1-8B-Instruct
Quantization
Q4_K_S
Framework
llama.cpp
Version
v0.34
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 M2 Ultra
Measured VRAM (avg)
15.2 GB
OS
macOS 26.1
Power draw
142.6 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

    v0.34 · macOS 26.1 · 8192 ctx

    123.15 tok/s

    Decode

    540.31 tok/s

    Prefill

    2.07 s

    TTFT

    15.2 GB

    VRAM

    —

    MTP

    142.6 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=103 running model_id=68 at Q4_K_S