Is 32GB RAM a good upgrade for running local LLMs like Llama 2 7B?

Yes. 32GB RAM gives you breathing room for a 4-bit quantized 7B model (5–6GB), your OS, apps, and a long context window without swapping to disk. 16GB works but chokes on decent chat lengths. RAM helps loading and context, but inference speed depends on your GPU or CPU compute—a GPU with 8–12GB VRAM is more impactful than system RAM.