Arao AI
Back to search results

Qwen3.5-27B · Q4_K_M

Start a discussion
NVIDIA RTX 3060 12GB12GBllama.cpp b4213L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-27B (Q4_K_M) on NVIDIA RTX 3060 12GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

45 tok/s

Decode

Decode speed

539 tok/s

Prefill

Prefill speed

1.15 s

TTFT

Time to first token

— GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ4_XS Decode 51 · Prefill 607 ·
  2. llama.cpp · Q4_K_M (current)Decode 45 · Prefill 539 ·
  3. llama.cpp · Q5_K_M Decode 51 · Prefill 535 ·
  4. llama.cpp · Q8_0 Decode 38 · Prefill 354 ·
  5. vLLM · IQ4_XS Decode 49 · Prefill 543 ·
  6. vLLM · Q4_K_M Decode 45 · Prefill 486 ·
  7. vLLM · Q5_K_M Decode 44 · Prefill 393 ·
  8. vLLM · Q8_0 Decode 28 · Prefill 330 ·
  9. Ollama · IQ4_XS Decode 57 · Prefill 462 ·
  10. Ollama · Q4_K_M Decode 52 · Prefill 443 ·
  11. Ollama · Q5_K_M Decode 43 · Prefill 500 ·
  12. Ollama · Q8_0 Decode 34 · Prefill 385 ·
  13. Strata · IQ4_XS Decode 77 · Prefill 864 ·
  14. Strata · Q4_K_M Decode 69 · Prefill 790 ·
  15. Strata · Q5_K_M Decode 67 · Prefill 774 ·
  16. Strata · Q8_0 Decode 54 · Prefill 449 ·

Core figures

Decode (avg)
45 tok/s
Prefill (avg)
539 tok/s
TTFT (avg)
1.15 s
VRAM (avg)
— GB
MTP acceptance rate
—
TTFB
— GB
Power draw
482 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-27B
Quantization
Q4_K_M
Framework
llama.cpp
Version
b4213
Context length
8192 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
OS
Linux 6.8.0
Driver
570.86.15
CUDA
12.6
Power draw
482 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-09

    b4213 · Linux 6.8.0 · CUDA 12.6 · 8192 ctx

    45 tok/s

    Decode

    539 tok/s

    Prefill

    1.15 s

    TTFT

    0 GB

    VRAM

    —

    MTP

    482 W

    Power draw

    Auto-generated LLM-sourced record 2026-10-09. Model: Qwen3.5-27B, Hardware: RTX 3060 12GB, Framework: 1, Quant: Q4_K_M.