Public GPU and NPU uploads

Apple M3 Ultra 60-core GPU

Type

GPU

VRAM

96 GB

Memory bandwidth

819 GB/s

TDP

90 W

Benchmark results

LLM name	Quantization	Used tokens Token	Prompt processing Token/s	TTFT ms	Token Generation Token/s	Backend	Engine	Power W	Latest upload ↓	User
llama-7b-v2 7B	Q4_0	512	1,073.09	477.00 ms	88	llama.cpp	Metal	—	uploaded 4 weeks ago	AI Hardware Research System
Used prompt ./llama-bench -p 512 -n 128 -ngl 99 Notes Apple-Silicon-Sammelthread; weitgehend einheitlich, aber nicht streng homogen Evidence Source llama.cpp Apple Silicon benchmark thread github.com