Arao AI
Back to search results

gemma-4-E2B-it · IQ3_XS

Start a discussion
NVIDIA DGX Spark128GBvLLM v0.50L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-E2B-it (IQ3_XS) on NVIDIA DGX Spark with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

360.01 tok/s

Decode

Decode speed

1,436.84 tok/s

Prefill

Prefill speed

1.75 s

TTFT

Time to first token

4.7 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. vLLM · IQ2_M Decode 380.91 · Prefill 2066.12 ·
  2. vLLM · IQ3_XS (current)Decode 360.01 · Prefill 1436.84 ·
  3. Strata · Q3_K_L Decode 389.31 · Prefill 1363.67 ·
  4. Strata · Q4_K_M Decode 326.71 · Prefill 1358.52 ·

Core figures

Decode (avg)
360.01 tok/s
Prefill (avg)
1,436.84 tok/s
TTFT (avg)
1.75 s
VRAM (avg)
4.7 GB
MTP acceptance rate
—
TTFB
2.88 GB
Power draw
249.3 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-E2B-it
Quantization
IQ3_XS
Framework
vLLM
Version
v0.50
Context length
32768 tokens
Batch size
32
GPU layers
999
Flash Attention
Off

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)
4.7 GB
OS
Windows 11 23H2
Driver
560.94
CUDA
12.8
Power draw
249.3 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

    v0.50 · Windows 11 23H2 · CUDA 12.8 · 32768 ctx

    360.01 tok/s

    Decode

    1,436.84 tok/s

    Prefill

    1.75 s

    TTFT

    4.7 GB

    VRAM

    —

    MTP

    249.3 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=54 running model_id=66 at IQ3_XS