Arao AI
Back to search results

Qwen3.5-4B · IQ3_S

Start a discussion
NVIDIA H100 80GB HBM380GBvLLM v0.51L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ3_S) on NVIDIA H100 80GB HBM3 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

329.61 tok/s

Decode

Decode speed

1,536.04 tok/s

Prefill

Prefill speed

1.13 s

TTFT

Time to first token

7.64 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. vLLM · IQ3_S (current)Decode 329.61 · Prefill 1536.04 ·
  2. vLLM · Q3_K_S Decode 362.27 · Prefill 1696.48 ·
  3. Strata · Q2_K Decode 373.13 · Prefill 1907.75 ·

Core figures

Decode (avg)
329.61 tok/s
Prefill (avg)
1,536.04 tok/s
TTFT (avg)
1.13 s
VRAM (avg)
7.64 GB
MTP acceptance rate
—
TTFB
2.64 GB
Power draw
648.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ3_S
Framework
vLLM
Version
v0.51
Context length
4096 tokens
Batch size
8
GPU layers
999
Flash Attention
Off

Hardware

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

GPU
NVIDIA H100 80GB HBM3
Nominal VRAM
80 GB
Measured VRAM (avg)
7.64 GB
OS
Ubuntu 24.04
Driver
570.86
CUDA
13.0
Power draw
648.9 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

    v0.51 · Ubuntu 24.04 · CUDA 13.0 · 4096 ctx

    329.61 tok/s

    Decode

    1,536.04 tok/s

    Prefill

    1.13 s

    TTFT

    7.64 GB

    VRAM

    —

    MTP

    648.9 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=55 running model_id=48 at IQ3_S