Qwen3.5-9B · Q5_K_S
This page aggregates 2 real-world runs of Qwen3.5-9B (Q5_K_S) on Apple M5 Max with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
163.05 tok/s
Decode
Decode speed
700.99 tok/s
Prefill
Prefill speed
0.86 s
TTFT
Time to first token
19.11 GB
VRAM
VRAM usage
2
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- vLLM · IQ3_S Decode 222.48 · Prefill 1168.2 ·
- vLLM · Q5_K_S (current)Decode 163.05 · Prefill 700.99 ·
- Strata · Q3_K_M Decode 284.27 · Prefill 948.35 ·
- vLLM · Q5_K_M Decode 148.01 · Prefill 738.05 ·
Core figures
- Decode (avg)
- 163.05 tok/s
- Prefill (avg)
- 700.99 tok/s
- TTFT (avg)
- 0.86 s
- VRAM (avg)
- 19.11 GB
- MTP acceptance rate
- —
- TTFB
- 1.31 GB
- Power draw
- 140.8 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-9B
- Quantization
- Q5_K_S
- Framework
- vLLM
- Version
- v0.33
- Context length
- 8192 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
- Apple M5 Max
- Nominal VRAM
- 128 GB
- Measured VRAM (avg)
- 19.11 GB
- OS
- Ubuntu 22.04
- Driver
- 560.94
- CUDA
- 12.6
- Power draw
- 174.2 W
Sources & evidence
2 measured records in total, each traceable to its original source
v0.33 · Ubuntu 22.04 · CUDA 12.6 · 8192 ctx
152.57 tok/s
Decode
518.94 tok/s
Prefill
0.41 s
TTFT
19.11 GB
VRAM
—
MTP
174.2 W
Power draw
L2 batch seed 2026-10-16: vLLM on hw_id=28 running model_id=49 at Q5_K_S
v0.28 · macOS 15.0 · CUDA N/A · 16384 ctx
173.52 tok/s
Decode
883.03 tok/s
Prefill
1.32 s
TTFT
19.11 GB
VRAM
—
MTP
107.4 W
Power draw
L2 batch seed 2026-10-13: vLLM on hw_id=28 running model_id=49 at Q5_K_S