Back to search results
Llama-3-8B-Instruct · IQ2_M
Apple M5 Ultra512GBllama.cpp v0.21L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3-8B-Instruct (IQ2_M) on Apple M5 Ultra with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
145.48 tok/s
Decode
Decode speed
672.26 tok/s
Prefill
Prefill speed
1.93 s
TTFT
Time to first token
10.84 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 (current)Decode 145.48 · Prefill 672.26 ·
- llama.cpp · Q4_K_S Decode 115.5 · Prefill 509.11 ·
- Ollama · IQ3_XS Decode 140.68 · Prefill 755.32 ·
Core figures
- Decode (avg)
- 145.48 tok/s
- Prefill (avg)
- 672.26 tok/s
- TTFT (avg)
- 1.93 s
- VRAM (avg)
- 10.84 GB
- MTP acceptance rate
- —
- TTFB
- 1.16 GB
- Power draw
- 127.2 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3-8B-Instruct
- Quantization
- IQ2_M
- Framework
- llama.cpp
- Version
- v0.21
- Context length
- 32768 tokens
- Batch size
- 256
- GPU layers
- 20
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- Apple M5 Ultra
- Nominal VRAM
- 512 GB
- Measured VRAM (avg)
- 10.84 GB
- OS
- macOS 26.1
- Power draw
- 127.2 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.21 · macOS 26.1 · 32768 ctx
145.48 tok/s
Decode
672.26 tok/s
Prefill
1.93 s
TTFT
10.84 GB
VRAM
—
MTP
127.2 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=29 running model_id=69 at IQ2_M