Back to search results
Llama-3-8B-Instruct · Q2_K
Apple M2 Ultrallama.cpp v0.31L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3-8B-Instruct (Q2_K) on Apple M2 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
151.53 tok/s
Decode
Decode speed
649.5 tok/s
Prefill
Prefill speed
1.41 s
TTFT
Time to first token
11.34 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
17 days ago
Last verified
Performance
- llama.cpp · IQ2_M Decode 144.03 · Prefill 634.04 ·
- llama.cpp · Q2_K (current)Decode 151.53 · Prefill 649.5 ·
- Ollama · IQ3_S Decode 116.94 · Prefill 440.9 ·
- Ollama · Q3_K_M Decode 114.21 · Prefill 572.36 ·
Core figures
- Decode (avg)
- 151.53 tok/s
- Prefill (avg)
- 649.5 tok/s
- TTFT (avg)
- 1.41 s
- VRAM (avg)
- 11.34 GB
- MTP acceptance rate
- —
- TTFB
- 1.21 GB
- Power draw
- 127.1 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3-8B-Instruct
- Quantization
- Q2_K
- Framework
- llama.cpp
- Version
- v0.31
- 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 M2 Ultra
- Measured VRAM (avg)
- 11.34 GB
- OS
- macOS 26.1
- Power draw
- 127.1 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.31 · macOS 26.1 · 2048 ctx
151.53 tok/s
Decode
649.5 tok/s
Prefill
1.41 s
TTFT
11.34 GB
VRAM
—
MTP
127.1 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=103 running model_id=69 at Q2_K