A Relatively recent gaming-type setup with local-ai or llama.cpp is what I’d recommend.
I do most of my AI stuff with an rtx3070, but I also have a ryzen 7 3800x with 64gb RAM for heavy models where I don’t so much care how long it takes but need the high parameter count for whatever reason, for example MoE and agentic behavior.
I’m not sure what you mean by ideal. Like, run any model you ever wanted? Probably the latest ai nvidia chips.
But you can get away with a lot less for smaller models. I have the amd mid range card from 4 years ago (i forget the model at the top of my head) and can run text, 8B sized, models without issue.
Tbf im not sure how much it helps them if you’re using the LLM without an account
Market share. They can show the usage figures to investors and ask for more cash
Not if you run it locally!
So what’s the ideal setup then?
A Relatively recent gaming-type setup with local-ai or llama.cpp is what I’d recommend.
I do most of my AI stuff with an rtx3070, but I also have a ryzen 7 3800x with 64gb RAM for heavy models where I don’t so much care how long it takes but need the high parameter count for whatever reason, for example MoE and agentic behavior.
I’m not sure what you mean by ideal. Like, run any model you ever wanted? Probably the latest ai nvidia chips.
But you can get away with a lot less for smaller models. I have the amd mid range card from 4 years ago (i forget the model at the top of my head) and can run text, 8B sized, models without issue.
I’m sorry, I use chatgpt for writing mysql queries and dax-formulas so that would be the use case.