Arao AI
Back to search results

gemma-4-31B-it · IQ3_XS

Start a discussion
NVIDIA DGX Spark128GBllama.cpp v0.31L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-31B-it (IQ3_XS) on NVIDIA DGX Spark with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

91.61 tok/s

Decode

Decode speed

449.28 tok/s

Prefill

Prefill speed

2.03 s

TTFT

Time to first token

42.46 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. llama.cpp · IQ3_XS (current)Decode 91.61 · Prefill 449.28 ·
  2. Ollama · Q6_K Decode 50.14 · Prefill 188.84 ·

Core figures

Decode (avg)
91.61 tok/s
Prefill (avg)
449.28 tok/s
TTFT (avg)
2.03 s
VRAM (avg)
42.46 GB
MTP acceptance rate
—
TTFB
0.73 GB
Power draw
261.9 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-31B-it
Quantization
IQ3_XS
Framework
llama.cpp
Version
v0.31
Context length
32768 tokens
Batch size
512
GPU layers
28
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)
42.46 GB
OS
Ubuntu 24.04
Driver
555.85
CUDA
12.8
Power draw
261.9 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.31 · Ubuntu 24.04 · CUDA 12.8 · 32768 ctx

    91.61 tok/s

    Decode

    449.28 tok/s

    Prefill

    2.03 s

    TTFT

    42.46 GB

    VRAM

    —

    MTP

    261.9 W

    Power draw

    L2 batch seed 2026-10-13: llama.cpp on hw_id=54 running model_id=63 at IQ3_XS