Qwen3-4B · Q5_K_M
This page aggregates 2 real-world runs of Qwen3-4B (Q5_K_M) on NVIDIA DGX Spark with Strata, contributed by 1 independent source platforms; metrics are averages of published measurements.
221.01 tok/s
Decode
Decode speed
864.66 tok/s
Prefill
Prefill speed
1.24 s
TTFT
Time to first token
9.91 GB
VRAM
VRAM usage
2
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · Q3_K_M Decode 206.5 · Prefill 1076.24 ·
- Strata · Q3_K_S Decode 338.04 · Prefill 1544.38 ·
- Strata · Q5_K_M (current)Decode 221.01 · Prefill 864.66 ·
- vLLM · Q3_K_L Decode 202.71 · Prefill 852.86 ·
- Strata · IQ1_S Decode 323.83 · Prefill 1439.4 ·
Core figures
- Decode (avg)
- 221.01 tok/s
- Prefill (avg)
- 864.66 tok/s
- TTFT (avg)
- 1.24 s
- VRAM (avg)
- 9.91 GB
- MTP acceptance rate
- 0.67
- TTFB
- 1.77 GB
- Power draw
- 254.95 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B
- Quantization
- Q5_K_M
- Framework
- Strata
- Version
- v0.21
- Context length
- 4096 tokens
- Batch size
- 512
- GPU layers
- 12
- 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)
- 9.91 GB
- OS
- macOS 15.0
- Driver
- 570.86
- CUDA
- 12.4
- Power draw
- 266.2 W
Sources & evidence
2 measured records in total, each traceable to its original source
v0.21 · macOS 15.0 · CUDA 12.4 · 4096 ctx
221.82 tok/s
Decode
760.91 tok/s
Prefill
1.45 s
TTFT
9.91 GB
VRAM
0.62
MTP
266.2 W
Power draw
L2 batch seed 2026-10-16: Strata on hw_id=54 running model_id=54 at Q5_K_M
v0.31 · Ubuntu 24.04 · CUDA 12.8 · 8192 ctx
220.2 tok/s
Decode
968.4 tok/s
Prefill
1.04 s
TTFT
9.91 GB
VRAM
0.72
MTP
243.7 W
Power draw
L2 batch seed 2026-10-13: Strata on hw_id=54 running model_id=54 at Q5_K_M