Arao AI
Back to search results

Qwen3.5-4B · IQ4_XS

Start a discussion
NVIDIA RTX 409024GBllama.cpp v0.25L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ4_XS) on NVIDIA RTX 4090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

170.35 tok/s

Decode

Decode speed

636.3 tok/s

Prefill

Prefill speed

0.3 s

TTFT

Time to first token

8.32 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · FP16 Decode 79.5 · Prefill 355.19 ·
  2. llama.cpp · IQ4_XS (current)Decode 170.35 · Prefill 636.3 ·
  3. vLLM · IQ2_M Decode 248.85 · Prefill 1225.63 ·
  4. vLLM · Q6_K Decode 133.61 · Prefill 629.89 ·
  5. llama.cpp · Q3_K_L Decode 174.81 · Prefill 809.7 ·
  6. vLLM · Q4_K_S Decode 181.53 · Prefill 809.47 ·
  7. Strata · Q6_K Decode 161.44 · Prefill 654.89 ·

Core figures

Decode (avg)
170.35 tok/s
Prefill (avg)
636.3 tok/s
TTFT (avg)
0.3 s
VRAM (avg)
8.32 GB
MTP acceptance rate
—
TTFB
1.36 GB
Power draw
325 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ4_XS
Framework
llama.cpp
Version
v0.25
Context length
2048 tokens
Batch size
256
GPU layers
99
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 4090
Nominal VRAM
24 GB
Measured VRAM (avg)
8.32 GB
OS
Ubuntu 24.04
Driver
565.57
CUDA
13.0
Power draw
325 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-18

    v0.25 · Ubuntu 24.04 · CUDA 13.0 · 2048 ctx

    170.35 tok/s

    Decode

    636.3 tok/s

    Prefill

    0.3 s

    TTFT

    8.32 GB

    VRAM

    —

    MTP

    325 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=1 running model_id=48 at IQ4_XS