Arao AI
Back to search results

Qwen3.5-4B · IQ3_S

Start a discussion
NVIDIA RTX 3060 12GB12GBllama.cpp v0.49L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ3_S) on NVIDIA RTX 3060 12GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

105.18 tok/s

Decode

Decode speed

415.05 tok/s

Prefill

Prefill speed

1.88 s

TTFT

Time to first token

7.64 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ3_S (current)Decode 105.18 · Prefill 415.05 ·
  2. Ollama · IQ3_S Decode 95.65 · Prefill 519.99 ·
  3. Ollama · Q3_K_M Decode 97.19 · Prefill 407.59 ·
  4. Strata · Q5_K_S Decode 113.91 · Prefill 618.95 ·
  5. Strata · Q6_K Decode 100.84 · Prefill 413.9 ·
  6. llama.cpp · Q4_K_M Decode 117 · Prefill 877.5 ·

Core figures

Decode (avg)
105.18 tok/s
Prefill (avg)
415.05 tok/s
TTFT (avg)
1.88 s
VRAM (avg)
7.64 GB
MTP acceptance rate
—
TTFB
0.84 GB
Power draw
242.2 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.49
Context length
4096 tokens
Batch size
512
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 3060 12GB
Nominal VRAM
12 GB
Measured VRAM (avg)
7.64 GB
OS
Ubuntu 22.04
Driver
565.57
CUDA
12.4
Power draw
242.2 W

Sources & evidence

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

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

    v0.49 · Ubuntu 22.04 · CUDA 12.4 · 4096 ctx

    105.18 tok/s

    Decode

    415.05 tok/s

    Prefill

    1.88 s

    TTFT

    7.64 GB

    VRAM

    —

    MTP

    242.2 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=15 running model_id=48 at IQ3_S