Watch a conference talk from ElixirConf EU 2024 that delves into the performance aspects of the Nx project in Elixir's AI and machine learning ecosystem. Learn how Nx leverages the BEAM for ML workflows, enabling concurrent model execution across distributed nodes and multiple GPUs. Understand the intricate balance between NIF computations and GPU scheduling, exploring how "Servings" and "Batches" optimize request handling and resource utilization. Discover the trade-offs involved in request batching, scheduling mechanisms, and the various constraints encountered when operating ML services. Gain insights into the technical solutions that enable Nx to deliver high throughput while managing the complexities of NIF scheduling and GPU parallel processing limitations.