Community-tested recipes, compose configs, and benchmarks for serving LLMs locally on RTX 3090-class GPUs.