Arao AI
Back to search results

Qwen3-8B · FP16

Start a discussion
Apple M1 Prollama.cpp v0.55L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (FP16) on Apple M1 Pro with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

33.62 tok/s

Decode

Decode speed

122.03 tok/s

Prefill

Prefill speed

1.04 s

TTFT

Time to first token

30.4 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 · FP16 (current)Decode 33.62 · Prefill 122.03 ·
  2. llama.cpp · Q5_K_M Decode 60.3 · Prefill 277.54 ·
  3. Strata · Q5_K_M Decode 74.46 · Prefill 397.25 ·

Core figures

Decode (avg)
33.62 tok/s
Prefill (avg)
122.03 tok/s
TTFT (avg)
1.04 s
VRAM (avg)
30.4 GB
MTP acceptance rate
—
TTFB
0.27 GB
Power draw
42 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
FP16
Framework
llama.cpp
Version
v0.55
Context length
32768 tokens
Batch size
256
GPU layers
33
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
Apple M1 Pro
Measured VRAM (avg)
30.4 GB
OS
macOS 15.0
Driver
Apple M1 Pro
CUDA
N/A
Power draw
42 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

    v0.55 · macOS 15.0 · CUDA N/A · 32768 ctx

    33.62 tok/s

    Decode

    122.03 tok/s

    Prefill

    1.04 s

    TTFT

    30.4 GB

    VRAM

    —

    MTP

    42 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=100 running model_id=53 at FP16