Arao AI
Back to search results

Qwen3-8B · IQ4_XS

Start a discussion
NVIDIA DGX Spark128GBvLLM v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-8B (IQ4_XS) on NVIDIA DGX Spark with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

168.28 tok/s

Decode

Decode speed

770.17 tok/s

Prefill

Prefill speed

1.25 s

TTFT

Time to first token

14.53 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. vLLM · Q3_K_M Decode 167.2 · Prefill 861.79 ·
  2. Strata · Q5_K_S Decode 197.21 · Prefill 856.13 ·
  3. llama.cpp · Q4_K_M Decode 146.4 · Prefill 699.64 ·
  4. vLLM · IQ4_XS (current)Decode 168.28 · Prefill 770.17 ·

Core figures

Decode (avg)
168.28 tok/s
Prefill (avg)
770.17 tok/s
TTFT (avg)
1.25 s
VRAM (avg)
14.53 GB
MTP acceptance rate
—
TTFB
1.35 GB
Power draw
265.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-8B
Quantization
IQ4_XS
Framework
vLLM
Version
v0.52
Context length
2048 tokens
Batch size
32
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 DGX Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
14.53 GB
OS
Ubuntu 24.04
Driver
555.85
CUDA
12.8
Power draw
265.8 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.52 · Ubuntu 24.04 · CUDA 12.8 · 2048 ctx

    168.28 tok/s

    Decode

    770.17 tok/s

    Prefill

    1.25 s

    TTFT

    14.53 GB

    VRAM

    —

    MTP

    265.8 W

    Power draw

    L2 batch seed 2026-10-13: vLLM on hw_id=54 running model_id=53 at IQ4_XS