Track manipulation, locomotion and embodied systems.
Search papers, labs, and topics across Lattice.
This work benchmarks inference backends for machine learning models used in fusion reactor control, revealing significant latency differences between CPU and GPU execution. For models exceeding five million parameters, CPU inference times range from tens to thousands of milliseconds, rendering them unsuitable for real-time control loops. The study emphasizes that the choice of deployment backend is intrinsically linked to model selection and the stringent timing requirements of fusion control systems.
Fusion reactor control ML models are too slow on CPUs for real-time operation, with GPUs offering a critical speedup for models over 5M parameters.
Machine learning models are increasingly used in feedback control loops for nuclear fusion, where inference speed and predictable timing are critical. We summarize lessons from models deployed for control on the DIII-D tokamak and develop a benchmark to compare inference backends across ten neural networks and model components from fusion control and diagnostic pipelines. For models greater than five million parameters, the CPU backends take tens to thousands of milliseconds, while GPU inference is substantially faster, suggesting an upper limit on CPU-oriented development for control. These results show why the deployment backend must be selected together with the model and its control-cycle budget.