Arao AI
Back to search results

Qwen3.6-35B-A3B · Q3_K_S

Start a discussion
NVIDIA H2097GBvLLM v0.21L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.6-35B-A3B (Q3_K_S) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

118.48 tok/s

Decode

Decode speed

535.48 tok/s

Prefill

Prefill speed

2.27 s

TTFT

Time to first token

49.88 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 118.48 · Prefill 535.48 ·
  2. vLLM · Q8_0 Decode 66.61 · Prefill 237.96 ·
  3. Ollama · Q3_K_M Decode 90.2 · Prefill 441.65 ·

Core figures

Decode (avg)
118.48 tok/s
Prefill (avg)
535.48 tok/s
TTFT (avg)
2.27 s
VRAM (avg)
49.88 GB
MTP acceptance rate
—
TTFB
0.95 GB
Power draw
321.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.6-35B-A3B
Quantization
Q3_K_S
Framework
vLLM
Version
v0.21
Context length
8192 tokens
Batch size
8
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)
49.88 GB
OS
Windows 11 23H2
Driver
570.86
CUDA
12.8
Power draw
321.9 W

Sources & evidence

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

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

    v0.21 · Windows 11 23H2 · CUDA 12.8 · 8192 ctx

    118.48 tok/s

    Decode

    535.48 tok/s

    Prefill

    2.27 s

    TTFT

    49.88 GB

    VRAM

    —

    MTP

    321.9 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=56 running model_id=47 at Q3_K_S