Back to search results
Qwen3.5-9B · FP16
Apple M2 UltravLLM v0.54L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-9B (FP16) on Apple M2 Ultra with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
54.48 tok/s
Decode
Decode speed
205.02 tok/s
Prefill
Prefill speed
0.3 s
TTFT
Time to first token
37.06 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
2 days ago
Last verified
Performance
- llama.cpp · Q3_K_S Decode 108.42 · Prefill 589.72 ·
- vLLM · FP16 (current)Decode 54.48 · Prefill 205.02 ·
- vLLM · Q3_K_L Decode 115.94 · Prefill 437.41 ·
Core figures
- Decode (avg)
- 54.48 tok/s
- Prefill (avg)
- 205.02 tok/s
- TTFT (avg)
- 0.3 s
- VRAM (avg)
- 37.06 GB
- MTP acceptance rate
- —
- TTFB
- 0.44 GB
- Power draw
- 24.3 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-9B
- Quantization
- FP16
- Framework
- vLLM
- Version
- v0.54
- Context length
- 32768 tokens
- Batch size
- 32
- 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
- Apple M2 Ultra
- Measured VRAM (avg)
- 37.06 GB
- OS
- macOS 15.0
- Driver
- Apple M2 Ultra
- CUDA
- N/A
- Power draw
- 24.3 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.54 · macOS 15.0 · CUDA N/A · 32768 ctx
54.48 tok/s
Decode
205.02 tok/s
Prefill
0.3 s
TTFT
37.06 GB
VRAM
—
MTP
24.3 W
Power draw
L2 batch seed 2026-10-16: vLLM on hw_id=103 running model_id=49 at FP16