Arao AI
Back to search results

gemma-4-12B-it · Q4_K_M

Start a discussion
NVIDIA RTX Spark128GBllama.cpp v0.36L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-12B-it (Q4_K_M) on NVIDIA RTX Spark with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

39.59 tok/s

Decode

Decode speed

158.74 tok/s

Prefill

Prefill speed

0.26 s

TTFT

Time to first token

22.76 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_L Decode 43.89 · Prefill 217.76 ·
  2. llama.cpp · Q4_K_M (current)Decode 39.59 · Prefill 158.74 ·
  3. Ollama · FP16 Decode 18.58 · Prefill 81.79 ·
  4. Ollama · Q4_K_M Decode 36.57 · Prefill 126.77 ·

Core figures

Decode (avg)
39.59 tok/s
Prefill (avg)
158.74 tok/s
TTFT (avg)
0.26 s
VRAM (avg)
22.76 GB
MTP acceptance rate
—
TTFB
0.32 GB
Power draw
66.9 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
Q4_K_M
Framework
llama.cpp
Version
v0.36
Context length
8192 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
NVIDIA RTX Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
22.76 GB
OS
Windows 11 24H2
Driver
NVIDIA 590.10
CUDA
13.0
Power draw
66.9 W
Reference price
$2,399 · updated 2026-10-19

Prices are not live; check the store page

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.36 · Windows 11 24H2 · CUDA 13.0 · 8192 ctx

    39.59 tok/s

    Decode

    158.74 tok/s

    Prefill

    0.26 s

    TTFT

    22.76 GB

    VRAM

    —

    MTP

    66.9 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=110 running model_id=65 at Q4_K_M