Arao AI
Back to search results

Qwen3.8-9B · IQ3_S

Start a discussion
NVIDIA H2097GBvLLM v0.54L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.8-9B (IQ3_S) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

57.73 tok/s

Decode

Decode speed

250.91 tok/s

Prefill

Prefill speed

1.7 s

TTFT

Time to first token

14.57 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 · IQ3_S (current)Decode 57.73 · Prefill 250.91 ·
  2. vLLM · Q3_K_L Decode 54.56 · Prefill 274.28 ·
  3. Strata · IQ3_S Decode 80.72 · Prefill 287.96 ·
  4. Strata · Q3_K_M Decode 74.4 · Prefill 349.95 ·

Core figures

Decode (avg)
57.73 tok/s
Prefill (avg)
250.91 tok/s
TTFT (avg)
1.7 s
VRAM (avg)
14.57 GB
MTP acceptance rate
—
TTFB
0.46 GB
Power draw
394.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.8-9B
Quantization
IQ3_S
Framework
vLLM
Version
v0.54
Context length
32768 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)
14.57 GB
OS
Ubuntu 24.04
Driver
572.83
CUDA
12.6
Power draw
394.8 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

    v0.54 · Ubuntu 24.04 · CUDA 12.6 · 32768 ctx

    57.73 tok/s

    Decode

    250.91 tok/s

    Prefill

    1.7 s

    TTFT

    14.57 GB

    VRAM

    —

    MTP

    394.8 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=56 running model_id=52 at IQ3_S