Arao AI
Back to search results

Qwen3-4B · FP16

Start a discussion
Apple M2 UltraOllama v0.55L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B (FP16) on Apple M2 Ultra with Ollama, contributed by 1 independent source platforms; metrics are averages of published measurements.

64.08 tok/s

Decode

Decode speed

216.32 tok/s

Prefill

Prefill speed

2.28 s

TTFT

Time to first token

17.64 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. Ollama · FP16 (current)Decode 64.08 · Prefill 216.32 ·
  2. Ollama · IQ2_XS Decode 161.88 · Prefill 564.86 ·
  3. Strata · Q3_K_L Decode 200.62 · Prefill 720.62 ·
  4. Strata · Q5_K_S Decode 162.1 · Prefill 686.85 ·

Core figures

Decode (avg)
64.08 tok/s
Prefill (avg)
216.32 tok/s
TTFT (avg)
2.28 s
VRAM (avg)
17.64 GB
MTP acceptance rate
—
TTFB
0.51 GB
Power draw
35 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B
Quantization
FP16
Framework
Ollama
Version
v0.55
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)
17.64 GB
OS
macOS 15.0
Driver
Apple M2 Ultra
CUDA
N/A
Power draw
35 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 · 16384 ctx

    64.08 tok/s

    Decode

    216.32 tok/s

    Prefill

    2.28 s

    TTFT

    17.64 GB

    VRAM

    —

    MTP

    35 W

    Power draw

    L2 batch seed 2026-10-16: Ollama on hw_id=103 running model_id=54 at FP16