Arao AI
Back to search results

Qwen3.5-0.8B · IQ2_XS

Start a discussion
NVIDIA DGX Spark128GBvLLM v0.30L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-0.8B (IQ2_XS) on NVIDIA DGX Spark with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

310.29 tok/s

Decode

Decode speed

1,000.44 tok/s

Prefill

Prefill speed

2.38 s

TTFT

Time to first token

2.91 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. vLLM · IQ2_XS (current)Decode 310.29 · Prefill 1000.44 ·
  2. Ollama · IQ2_XS Decode 348.98 · Prefill 1689.57 ·
  3. Ollama · Q3_K_M Decode 387.92 · Prefill 1612.07 ·
  4. Ollama · IQ3_XS Decode 391.28 · Prefill 1950.82 ·
  5. Ollama · IQ4_XS Decode 338.64 · Prefill 1719.5 ·

Core figures

Decode (avg)
310.29 tok/s
Prefill (avg)
1,000.44 tok/s
TTFT (avg)
2.38 s
VRAM (avg)
2.91 GB
MTP acceptance rate
—
TTFB
2.48 GB
Power draw
193.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-0.8B
Quantization
IQ2_XS
Framework
vLLM
Version
v0.30
Context length
8192 tokens
Batch size
16
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 DGX Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
2.91 GB
OS
Windows 11 23H2
Driver
560.94
CUDA
12.6
Power draw
193.1 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.30 · Windows 11 23H2 · CUDA 12.6 · 8192 ctx

    310.29 tok/s

    Decode

    1,000.44 tok/s

    Prefill

    2.38 s

    TTFT

    2.91 GB

    VRAM

    —

    MTP

    193.1 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=54 running model_id=51 at IQ2_XS