16GB RAM vs 32GB RAM for Llama 7B and 13B: 32GB is the sweet spot for real use

7B runs okay on 16GB if you close everything else, but 13B demands 32GB unless you enjoy waiting forever. The math is simple: 7B in 4-bit needs 4–6GB for weights, plus OS and apps, so 16GB is borderline. 13B needs 8–10GB for weights alone, pushing 16GB to its limit. 32GB gives you headroom for faster inference, longer prompts, and higher precision—Future You will thank you.