Arao AI
Back to search results

Qwen3.5-27B · IQ3_XS

Start a discussion
Apple M5 Max128GBllama.cpp v0.33L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (IQ3_XS) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

171.72 tok/s

Decode

Decode speed

594.08 tok/s

Prefill

Prefill speed

1.23 s

TTFT

Time to first token

37.25 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 · IQ3_XS (current)Decode 171.72 · Prefill 594.08 ·
  2. llama.cpp · Q4_K_S Decode 146.98 · Prefill 610.13 ·
  3. Ollama · Q5_K_M Decode 119.42 · Prefill 542.35 ·
  4. Ollama · Q6_K Decode 100.18 · Prefill 374.17 ·

Core figures

Decode (avg)
171.72 tok/s
Prefill (avg)
594.08 tok/s
TTFT (avg)
1.23 s
VRAM (avg)
37.25 GB
MTP acceptance rate
—
TTFB
1.37 GB
Power draw
163.4 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-27B
Quantization
IQ3_XS
Framework
llama.cpp
Version
v0.33
Context length
32768 tokens
Batch size
512
GPU layers
28
Flash Attention
On

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)
37.25 GB
OS
macOS 15.0
Driver
570.86
CUDA
12.6
Power draw
163.4 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

    v0.33 · macOS 15.0 · CUDA 12.6 · 32768 ctx

    171.72 tok/s

    Decode

    594.08 tok/s

    Prefill

    1.23 s

    TTFT

    37.25 GB

    VRAM

    —

    MTP

    163.4 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=28 running model_id=50 at IQ3_XS