Is 16GB of RAM enough for running Llama 3.1 8B at 4-bit?

Yes, 16GB works. The model takes about 5-6GB to load, leaving 9-10GB free for inference, context, and OS overhead—enough to run the model and a browser. 8GB is a hard no: you’ll have maybe 2GB free after loading, causing heavy swapping and borderline unusable performance. For that model, 16GB is the minimum for comfortable use.