Back to search results
Qwen3.5-4B · Q3_K_S
NVIDIA RTX 409024GBllama.cpp v0.32L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-4B (Q3_K_S) on NVIDIA RTX 4090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
181.19 tok/s
Decode
Decode speed
930.71 tok/s
Prefill
Prefill speed
1.77 s
TTFT
Time to first token
7.56 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
17 days ago
Last verified
Performance
- llama.cpp · Q3_K_S (current)Decode 181.19 · Prefill 930.71 ·
- Strata · Q3_K_L Decode 236.63 · Prefill 1225.73 ·
- llama.cpp · FP16 Decode 79.5 · Prefill 355.19 ·
- llama.cpp · IQ4_XS Decode 170.35 · Prefill 636.3 ·
- vLLM · IQ2_M Decode 248.85 · Prefill 1225.63 ·
- vLLM · Q6_K Decode 133.61 · Prefill 629.89 ·
- llama.cpp · Q3_K_L Decode 174.81 · Prefill 809.7 ·
- vLLM · Q4_K_S Decode 181.53 · Prefill 809.47 ·
- Strata · Q6_K Decode 161.44 · Prefill 654.89 ·
Core figures
- Decode (avg)
- 181.19 tok/s
- Prefill (avg)
- 930.71 tok/s
- TTFT (avg)
- 1.77 s
- VRAM (avg)
- 7.56 GB
- MTP acceptance rate
- —
- TTFB
- 1.45 GB
- Power draw
- 362.8 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-4B
- Quantization
- Q3_K_S
- Framework
- llama.cpp
- Version
- v0.32
- Context length
- 32768 tokens
- Batch size
- 512
- GPU layers
- 33
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 4090
- Nominal VRAM
- 24 GB
- Measured VRAM (avg)
- 7.56 GB
- OS
- Windows 11 23H2
- Driver
- 560.94
- CUDA
- 12.8
- Power draw
- 362.8 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.32 · Windows 11 23H2 · CUDA 12.8 · 32768 ctx
181.19 tok/s
Decode
930.71 tok/s
Prefill
1.77 s
TTFT
7.56 GB
VRAM
—
MTP
362.8 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=1 running model_id=48 at Q3_K_S