Back to search results
Llama-3-8B-Instruct · Q3_K_M
Apple M5 Max128GBllama.cpp v0.50L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3-8B-Instruct (Q3_K_M) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
83.59 tok/s
Decode
Decode speed
361.92 tok/s
Prefill
Prefill speed
1.04 s
TTFT
Time to first token
13.52 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 · Q3_K_M (current)Decode 83.59 · Prefill 361.92 ·
- Ollama · IQ2_M Decode 100.02 · Prefill 413.11 ·
Core figures
- Decode (avg)
- 83.59 tok/s
- Prefill (avg)
- 361.92 tok/s
- TTFT (avg)
- 1.04 s
- VRAM (avg)
- 13.52 GB
- MTP acceptance rate
- —
- TTFB
- 0.67 GB
- Power draw
- 61.5 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3-8B-Instruct
- Quantization
- Q3_K_M
- Framework
- llama.cpp
- Version
- v0.50
- Context length
- 4096 tokens
- Batch size
- 256
- GPU layers
- 99
- 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)
- 13.52 GB
- OS
- macOS 26.1
- Power draw
- 61.5 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.50 · macOS 26.1 · 4096 ctx
83.59 tok/s
Decode
361.92 tok/s
Prefill
1.04 s
TTFT
13.52 GB
VRAM
—
MTP
61.5 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=69 at Q3_K_M