Test your llama.cpp '--threads' argument for up to 80% performance gain

A Reddit user reports up to 80% performance improvement on llama.cpp by optimizing the --threads argument for hybrid CPU architectures. Suggests using only P-cores with taskset/affinity for best results.