Arao AI
Back to search results

Llama-3.2-3B-Instruct · IQ3_S

Start a discussion
NVIDIA RTX 3060 12GB12GBllama.cpp v0.55L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (IQ3_S) on NVIDIA RTX 3060 12GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

193.44 tok/s

Decode

Decode speed

733.2 tok/s

Prefill

Prefill speed

0.65 s

TTFT

Time to first token

6.26 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 193.44 · Prefill 733.2 ·
  2. llama.cpp · Q5_K_S Decode 126.84 · Prefill 630.8 ·
  3. vLLM · Q5_K_M Decode 143.38 · Prefill 521.02 ·
  4. llama.cpp · Q8_0 Decode 87.8 · Prefill 658.5 ·
  5. llama.cpp · Q5_K_M Decode 127.6 · Prefill 957 ·
  6. llama.cpp · Q4_K_M Decode 156 · Prefill 1170 ·

Core figures

Decode (avg)
193.44 tok/s
Prefill (avg)
733.2 tok/s
TTFT (avg)
0.65 s
VRAM (avg)
6.26 GB
MTP acceptance rate
—
TTFB
1.55 GB
Power draw
147.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3.2-3B-Instruct
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.55
Context length
16384 tokens
Batch size
512
GPU layers
99
Flash Attention
Off

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)
6.26 GB
OS
Windows 11 23H2
Driver
565.57
CUDA
12.6
Power draw
147.1 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

    v0.55 · Windows 11 23H2 · CUDA 12.6 · 16384 ctx

    193.44 tok/s

    Decode

    733.2 tok/s

    Prefill

    0.65 s

    TTFT

    6.26 GB

    VRAM

    —

    MTP

    147.1 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=15 running model_id=70 at IQ3_S