Arao AI
Back to search results

Qwen3.6-27B · Q3_K_S

Start a discussion
NVIDIA H2097GBvLLM v0.22L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.6-27B (Q3_K_S) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

48.48 tok/s

Decode

Decode speed

199.47 tok/s

Prefill

Prefill speed

1.89 s

TTFT

Time to first token

38.96 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. vLLM · Q3_K_S (current)Decode 48.48 · Prefill 199.47 ·
  2. Ollama · Q4_K_M Decode 32.97 · Prefill 115.23 ·

Core figures

Decode (avg)
48.48 tok/s
Prefill (avg)
199.47 tok/s
TTFT (avg)
1.89 s
VRAM (avg)
38.96 GB
MTP acceptance rate
—
TTFB
0.39 GB
Power draw
341 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.6-27B
Quantization
Q3_K_S
Framework
vLLM
Version
v0.22
Context length
8192 tokens
Batch size
16
GPU layers
999
Flash Attention
On

Hardware

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

GPU
NVIDIA H20
Nominal VRAM
97 GB
Measured VRAM (avg)
38.96 GB
OS
Windows 11 23H2
Driver
570.86
CUDA
12.6
Power draw
341 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.22 · Windows 11 23H2 · CUDA 12.6 · 8192 ctx

    48.48 tok/s

    Decode

    199.47 tok/s

    Prefill

    1.89 s

    TTFT

    38.96 GB

    VRAM

    —

    MTP

    341 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=56 running model_id=6 at Q3_K_S