Arao AI
Back to search results

gemma-4-E2B-it · Q3_K_S

Start a discussion
2x AMD Strix HalovLLM v0.40L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-E2B-it (Q3_K_S) on 2x AMD Strix Halo with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

277.77 tok/s

Decode

Decode speed

1,024.78 tok/s

Prefill

Prefill speed

2.05 s

TTFT

Time to first token

4.83 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 · Q3_K_S (current)Decode 277.77 · Prefill 1024.78 ·
  2. Strata · IQ3_XS Decode 321.59 · Prefill 1311.64 ·

Core figures

Decode (avg)
277.77 tok/s
Prefill (avg)
1,024.78 tok/s
TTFT (avg)
2.05 s
VRAM (avg)
4.83 GB
MTP acceptance rate
—
TTFB
2.22 GB
Power draw
66.4 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-E2B-it
Quantization
Q3_K_S
Framework
vLLM
Version
v0.40
Context length
2048 tokens
Batch size
8
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
2x AMD Strix Halo
Measured VRAM (avg)
4.83 GB
OS
Ubuntu 22.04
Driver
amdgpu 6.10
CUDA
N/A
Power draw
66.4 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

    v0.40 · Ubuntu 22.04 · CUDA N/A · 2048 ctx

    277.77 tok/s

    Decode

    1,024.78 tok/s

    Prefill

    2.05 s

    TTFT

    4.83 GB

    VRAM

    —

    MTP

    66.4 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=88 running model_id=66 at Q3_K_S