Is 32GB RAM enough for running large language models, or does VRAM matter more?

It depends on whether you’re running on CPU or GPU. For CPU inference (like llama.cpp or Ollama), 32GB is plenty—a 7B model in 4-bit uses about 4-6GB of system RAM. For GPU inference, VRAM is the bottleneck: a 7B model in 4-bit needs 6-8GB of VRAM, and full precision can use 14GB+.