Better GPU for ML?

Right now I'm choosing between RTX 4060 Ti 16GB and RTX 4070 Ghost 12GB (cost is exactly the same). What's better for machine learning and LLMs (and possibly physics simulations)? More VRAM sounds better as I would be able to host 7B LLM models without quantization, but with RTX 4070 I will have better performance (but on quantized models).

My additional reason for buying GPU is gaming, and that's where RTX 4070 shines.

I am also open to other options …

cost datascience ghost gpu llm llm models llms machine machine learning performance physics quantization rtx rtx 4060 rtx 4060 ti rtx 4070 simulations will