Back to search results
Qwen3.5-0.8B · IQ3_S
Apple M5 Max128GBllama.cpp v0.51L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3.5-0.8B (IQ3_S) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
184.05 tok/s
Decode
Decode speed
824.37 tok/s
Prefill
Prefill speed
1.86 s
TTFT
Time to first token
3.21 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · IQ3_S (current)Decode 184.05 · Prefill 824.37 ·
- Ollama · IQ3_S Decode 248.21 · Prefill 1111.71 ·
- Ollama · IQ4_XS Decode 157.93 · Prefill 586.06 ·
- Ollama · Q4_K_S Decode 308.8 · Prefill 1388.47 ·
- Ollama · Q5_K_M Decode 365.7 · Prefill 1467.25 ·
- Strata · FP16 Decode 330.86 · Prefill 1157.2 ·
- Strata · Q4_K_M Decode 357.61 · Prefill 1232.58 ·
- Ollama · Q2_K Decode 281.81 · Prefill 1519.14 ·
- Strata · IQ2_M Decode 308.09 · Prefill 1499.86 ·
Core figures
- Decode (avg)
- 184.05 tok/s
- Prefill (avg)
- 824.37 tok/s
- TTFT (avg)
- 1.86 s
- VRAM (avg)
- 3.21 GB
- MTP acceptance rate
- —
- TTFB
- 1.47 GB
- Power draw
- 45.9 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3.5-0.8B
- Quantization
- IQ3_S
- Framework
- llama.cpp
- Version
- v0.51
- Context length
- 2048 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
- Apple M5 Max
- Nominal VRAM
- 128 GB
- Measured VRAM (avg)
- 3.21 GB
- OS
- macOS 26.1
- Power draw
- 45.9 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.51 · macOS 26.1 · 2048 ctx
184.05 tok/s
Decode
824.37 tok/s
Prefill
1.86 s
TTFT
3.21 GB
VRAM
—
MTP
45.9 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=51 at IQ3_S