Exploring Neuromorphic Computing with Loihi-2 for High-Performance CFD Simulations

Modern computational fluid dynamics (CFD) simulations are reaching unprecedented scales, as seen in NASA’s X-59 effort involving 117 billion grid cells. Such complexity pushes conventional CPU and GPU architectures to their limits, particularly in energy efficiency and scalability for real-time turbulence modeling or adaptive grid refinement. This paper investigates Intel’s Loihi-2 neuromorphic processor as a potential solution. By relying on event-driven and asynchronous parallel processing, Loihi-2 significantly reduces power demands while managing high-throughput CFD workloads. Experimental results reveal a compute efficiency of 103.9 GFLOPs/W far exceeding the 3.8 GFLOPs/W of CPUs and 74.7 GFLOPs/W of GPUs. Per-iteration energy consumption drops to 0.09 Joules, representing an improvement of 34x over CPUs and 10x over GPUs. These performance gains underscore the practicality of neuromorphic hardware for energy-efficient, large-scale CFD, including turbulence modeling. Moreover, Loihi-2’s reduced power overhead positions it favorably for aerospace applications requiring onboard flow simulations, as well as real-time IoT scenarios. Overall, this study demonstrates the viability of integrating neuromorphic processors into next-generation CFD solvers, bridging neuroscience-inspired computation and high-performance engineering.