Arao AI
Back to search results

Kimi-K3 · Q3_K_M

Start a discussion
NVIDIA RTX 3060 12GB12GBllama.cpp v0.36L2 Cross-framework verified

This page aggregates 1 real-world runs of Kimi-K3 (Q3_K_M) on NVIDIA RTX 3060 12GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

58.85 tok/s

Decode

Decode speed

193.51 tok/s

Prefill

Prefill speed

1.17 s

TTFT

Time to first token

1.05 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q3_K_L Decode 55.93 · Prefill 219.18 ·
  2. llama.cpp · Q3_K_M (current)Decode 58.85 · Prefill 193.51 ·
  3. vLLM · IQ4_XS Decode 53.14 · Prefill 228.98 ·

Core figures

Decode (avg)
58.85 tok/s
Prefill (avg)
193.51 tok/s
TTFT (avg)
1.17 s
VRAM (avg)
1.05 GB
MTP acceptance rate
—
TTFB
0.47 GB
Power draw
160.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Kimi-K3
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.36
Context length
16384 tokens
Batch size
512
GPU layers
20
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)
1.05 GB
OS
Windows 11 23H2
Driver
560.94
CUDA
12.8
Power draw
160.7 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

    v0.36 · Windows 11 23H2 · CUDA 12.8 · 16384 ctx

    58.85 tok/s

    Decode

    193.51 tok/s

    Prefill

    1.17 s

    TTFT

    1.05 GB

    VRAM

    —

    MTP

    160.7 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=15 running model_id=73 at Q3_K_M