Arao AI
Back to search results

gemma-4-E4B-it · Q2_K

Start a discussion
NVIDIA DGX Spark128GBllama.cpp v0.37L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-E4B-it (Q2_K) on NVIDIA DGX Spark with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

258.21 tok/s

Decode

Decode speed

920.61 tok/s

Prefill

Prefill speed

0.49 s

TTFT

Time to first token

6.72 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 · Q2_K (current)Decode 258.21 · Prefill 920.61 ·
  2. vLLM · FP16 Decode 107.3 · Prefill 444.7 ·

Core figures

Decode (avg)
258.21 tok/s
Prefill (avg)
920.61 tok/s
TTFT (avg)
0.49 s
VRAM (avg)
6.72 GB
MTP acceptance rate
—
TTFB
2.07 GB
Power draw
243.1 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-E4B-it
Quantization
Q2_K
Framework
llama.cpp
Version
v0.37
Context length
32768 tokens
Batch size
512
GPU layers
20
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
NVIDIA DGX Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
6.72 GB
OS
Ubuntu 24.04
Driver
555.85
CUDA
12.8
Power draw
243.1 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

    v0.37 · Ubuntu 24.04 · CUDA 12.8 · 32768 ctx

    258.21 tok/s

    Decode

    920.61 tok/s

    Prefill

    0.49 s

    TTFT

    6.72 GB

    VRAM

    —

    MTP

    243.1 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=54 running model_id=67 at Q2_K