Back to search results
Qwen3-4B-Thinking-2507 · IQ1_S
NVIDIA DGX Spark128GBvLLM v0.33L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (IQ1_S) on NVIDIA DGX Spark with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
291.5 tok/s
Decode
Decode speed
1,360.04 tok/s
Prefill
Prefill speed
1.03 s
TTFT
Time to first token
5.63 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- vLLM · IQ1_S (current)Decode 291.5 · Prefill 1360.04 ·
- vLLM · Q5_K_M Decode 151.71 · Prefill 832.95 ·
- Strata · Q5_K_M Decode 220.61 · Prefill 780.86 ·
- Ollama · Q3_K_L Decode 191.76 · Prefill 876.65 ·
- Ollama · Q5_K_S Decode 148.12 · Prefill 634.82 ·
Core figures
- Decode (avg)
- 291.5 tok/s
- Prefill (avg)
- 1,360.04 tok/s
- TTFT (avg)
- 1.03 s
- VRAM (avg)
- 5.63 GB
- MTP acceptance rate
- —
- TTFB
- 2.33 GB
- Power draw
- 197.6 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B-Thinking-2507
- Quantization
- IQ1_S
- Framework
- vLLM
- Version
- v0.33
- Context length
- 2048 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 DGX Spark
- Nominal VRAM
- 128 GB
- Measured VRAM (avg)
- 5.63 GB
- OS
- Ubuntu 24.04
- Driver
- 565.57
- CUDA
- 12.4
- Power draw
- 197.6 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.33 · Ubuntu 24.04 · CUDA 12.4 · 2048 ctx
291.5 tok/s
Decode
1,360.04 tok/s
Prefill
1.03 s
TTFT
5.63 GB
VRAM
—
MTP
197.6 W
Power draw
L2 batch seed 2026-10-16: vLLM on hw_id=54 running model_id=57 at IQ1_S