I have 2 4090 and last time i played around with it, the context window killed it for me.

The normal LLMStudio stuff works great, but then i tried out anything with subagent things or parallel stuff and it trashed my cache and got super sluggish/slowish.

What do you run and how?