Fault-tolerant quantum computing (FTQC) at scale will require several classical computations to run in real time alongside the coherent control and readout of the QPU. These include decoding syndromes against many detector error models (DEMs), dynamically selecting the right DEM as the circuit evolves, and tracking drift in the underlying error probabilities. We will review some of NVIDIA’s recent work to understand these requirements and our plans for tools that help the community advance real-time accelerated computing for FTQC.