Arao AI
Back to search results

elephant-1 · Q3_K_S

Start a discussion
Apple M5 Ultra512GBllama.cpp v0.34L2 Cross-framework verified

This page aggregates 1 real-world runs of elephant-1 (Q3_K_S) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

117.38 tok/s

Decode

Decode speed

386.02 tok/s

Prefill

Prefill speed

2.43 s

TTFT

Time to first token

14.38 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

14 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_S (current)Decode 117.38 · Prefill 386.02 ·
  2. llama.cpp · Q5_K_S Decode 83.32 · Prefill 309.36 ·
  3. Ollama · Q6_K Decode 62.32 · Prefill 341.91 ·

Core figures

Decode (avg)
117.38 tok/s
Prefill (avg)
386.02 tok/s
TTFT (avg)
2.43 s
VRAM (avg)
14.38 GB
MTP acceptance rate
—
TTFB
0.94 GB
Power draw
125.4 W

Configuration

Member-level fields are taken from the most recent run

Model
elephant-1
Quantization
Q3_K_S
Framework
llama.cpp
Version
v0.34
Context length
8192 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 Ultra
Nominal VRAM
512 GB
Measured VRAM (avg)
14.38 GB
OS
macOS 26.1
Power draw
125.4 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-22

    v0.34 · macOS 26.1 · 8192 ctx

    117.38 tok/s

    Decode

    386.02 tok/s

    Prefill

    2.43 s

    TTFT

    14.38 GB

    VRAM

    —

    MTP

    125.4 W

    Power draw

    L2 batch seed 2026-10-22: llama.cpp on hw_id=29 running model_id=95 at Q3_K_S