How much VRAM does Q8 run on?

If you use Hermes and give it enough context, you don’t really need the top models at all in my experience

I wouldn't go that far. Billions of parameters vs trillions for starters. Q8 isn't even what the model is capable of.