Arao AI
Back to search results

mammoth-1 · Q3_K_M

Start a discussion
Apple M5 Ultra512GBllama.cpp v0.44L2 Cross-framework verified

This page aggregates 1 real-world runs of mammoth-1 (Q3_K_M) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

90.54 tok/s

Decode

Decode speed

497.72 tok/s

Prefill

Prefill speed

0.45 s

TTFT

Time to first token

40.66 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

13 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_M (current)Decode 90.54 · Prefill 497.72 ·
  2. llama.cpp · Q5_K_S Decode 71.27 · Prefill 325.67 ·
  3. Ollama · IQ4_XS Decode 81.44 · Prefill 327.07 ·

Core figures

Decode (avg)
90.54 tok/s
Prefill (avg)
497.72 tok/s
TTFT (avg)
0.45 s
VRAM (avg)
40.66 GB
MTP acceptance rate
—
TTFB
0.72 GB
Power draw
113.9 W

Configuration

Member-level fields are taken from the most recent run

Model
mammoth-1
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.44
Context length
8192 tokens
Batch size
256
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 M5 Ultra
Nominal VRAM
512 GB
Measured VRAM (avg)
40.66 GB
OS
macOS 26.1
Power draw
113.9 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

    v0.44 · macOS 26.1 · 8192 ctx

    90.54 tok/s

    Decode

    497.72 tok/s

    Prefill

    0.45 s

    TTFT

    40.66 GB

    VRAM

    —

    MTP

    113.9 W

    Power draw

    L2 batch seed 2026-10-22: llama.cpp on hw_id=29 running model_id=94 at Q3_K_M