Analyzing Energy Consumption of Loihi 2 Neuromorphic Chip in a Self-driving Use-case

A modern network often studied today, which comes closer to copying and imitating the dynamics and characteristics of a realistic brain, is the Spiking Neural Network (SNN). This network operates on completely new principles and requires specific hardware for efficient operation. In order to achieve these requirements, Intel researchers developed the Loihi 2 neuromorphic chip. In our research, we got to work with a Loihi 2 hardware system developed by Intel Labs. We explored the main differences in performance and accuracy compared to traditional GPUs. Using the hardware and provided framework, we carried out a comparison of applications implemented on both GPU and Loihi 2 hardware. The primary focus of the comparison was energy consumption and speed. The study involves benchmarking a vehicular steering prediction model implemented on both platforms, utilizing an SNN architecture optimized for neuromorphic computing. At the end of the study we conclude results of the implemented SNN on both hardware and show that Loihi 2 does not fall short of GPU regarding accuracy as it reaches a 0.89 R-squared score compared to the 0.94 R-squared score of the GPU, while only using 3 to 3.5 Watts on average in contrast to GPUs over 50 Watts power consumption on average.