With 32 GB RAM you can do inference with quantized 34b models. I wouldn’t call that useless?

You don’t need a GPU for llm inference. Might not be as fast as it could be but usable.