H100 80GB vs M2 16GB

For running local LLMs · 2 reports across 2 models

Side A

Vendor: nvidia
VRAM: 80GB
Memory: Discrete

Side B

Vendor: apple
VRAM: 16GB
Memory: Unified

Tokens per second by model

Model	H100 80GB	M2 16GB
Gemma 4up to 31B	—	18.0n=1
Qwen3.6up to 35B	45.0n=1	—

More comparisons