Arao AI
Back to search results

Qwen3-4B · IQ3_XS

Start a discussion
NVIDIA RTX 509032GBOllama v0.22L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B (IQ3_XS) on NVIDIA RTX 5090 with Ollama, contributed by 1 independent source platforms; metrics are averages of published measurements.

349.78 tok/s

Decode

Decode speed

1,805.86 tok/s

Prefill

Prefill speed

2.21 s

TTFT

Time to first token

7.31 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 · FP16 Decode 149.3 · Prefill 589.87 ·
  2. llama.cpp · IQ4_XS Decode 335.8 · Prefill 1082.78 ·
  3. Ollama · IQ3_XS (current)Decode 349.78 · Prefill 1805.86 ·

Core figures

Decode (avg)
349.78 tok/s
Prefill (avg)
1,805.86 tok/s
TTFT (avg)
2.21 s
VRAM (avg)
7.31 GB
MTP acceptance rate
—
TTFB
2.8 GB
Power draw
324.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B
Quantization
IQ3_XS
Framework
Ollama
Version
v0.22
Context length
2048 tokens
Batch size
256
GPU layers
33
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
7.31 GB
OS
Windows 11 23H2
Driver
570.86
CUDA
12.4
Power draw
324.6 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

    v0.22 · Windows 11 23H2 · CUDA 12.4 · 2048 ctx

    349.78 tok/s

    Decode

    1,805.86 tok/s

    Prefill

    2.21 s

    TTFT

    7.31 GB

    VRAM

    —

    MTP

    324.6 W

    Power draw

    L2 batch seed 2026-10-16: Ollama on hw_id=13 running model_id=54 at IQ3_XS