Let’s face it, GPUs are a scarce resource. As value much as LLMs can add to business in the Healthcare industry, we also need to be mindful of cost. In this talk, we will go through practical advices and suggestions on how to choose GPUs for RAG clinical note QA bot based on data size, text distribution, and model choices. We will also demonstrate how to inspect GPU performance and unveil the knobs that you can tune to improve it. This is used to create embeddings at billion scale and successfully unlock the potential of clinical notes for a healthcare provider.