Qwen3.5-4B · Q6_K
This page aggregates 2 real-world runs of Qwen3.5-4B (Q6_K) on AMD Ryzen AI Max+ 395 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
110.75 tok/s
Decode
Decode speed
500.4 tok/s
Prefill
Prefill speed
1.53 s
TTFT
Time to first token
10.5 GB
VRAM
VRAM usage
2
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · IQ3_XS Decode 156.15 · Prefill 562.48 ·
- vLLM · IQ2_XS Decode 205.56 · Prefill 924.99 ·
- vLLM · Q6_K (current)Decode 110.75 · Prefill 500.4 ·
Core figures
- Decode (avg)
- 110.75 tok/s
- Prefill (avg)
- 500.4 tok/s
- TTFT (avg)
- 1.53 s
- VRAM (avg)
- 10.5 GB
- MTP acceptance rate
- —
- TTFB
- 0.89 GB
- Power draw
- 73.75 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-4B
- Quantization
- Q6_K
- Framework
- vLLM
- Version
- v0.26
- Context length
- 32768 tokens
- Batch size
- 8
- 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
- AMD Ryzen AI Max+ 395
- Nominal VRAM
- 128 GB
- Measured VRAM (avg)
- 10.5 GB
- OS
- Ubuntu 24.04
- Driver
- amdgpu-pro 24.20
- CUDA
- N/A
- Power draw
- 112 W
Sources & evidence
2 measured records in total, each traceable to its original source
v0.26 · Ubuntu 24.04 · CUDA N/A · 32768 ctx
107.76 tok/s
Decode
383.23 tok/s
Prefill
1.11 s
TTFT
10.5 GB
VRAM
—
MTP
112 W
Power draw
L2 batch seed 2026-10-16: vLLM on hw_id=24 running model_id=48 at Q6_K
v0.27 · Windows 11 23H2 · CUDA N/A · 2048 ctx
113.73 tok/s
Decode
617.57 tok/s
Prefill
1.95 s
TTFT
10.5 GB
VRAM
—
MTP
35.5 W
Power draw
L2 batch seed 2026-10-13: vLLM on hw_id=24 running model_id=48 at Q6_K