Mapping GPUs to LLMs (and back): A bandwidth-based estimator for local inference

localllm-advisor.com